Rs. 395.00 CLASSIC DATA STRUCTURES, 2nd ed. (With CD-ROM) Debasis Samanta © 2009 by FHI Learning Private Limited, New Delhi. All rights reserved. No part of this book may be reproduced in any form, by mimeograph or any other means, without permission in writing from the publisher. Warning and Disclaimer The author and the publisher are not responsible for any loss or damage arising from the usa of CD-ROM included wth this book. The CD-ROM included with this book has no commercial value and cannot be sold coparately. ISBN-978-81-203-9731-2 The export rights of this book are vested solely with the publisher. Seventeenth Printing (Second Edition) ~ - July, 2009 Published by Asoke K. Ghosh, PHI Leaming Private Limited, M-97, Connaught Gircus, New Delhi-1 10001 and Printed by Rajkamal Electric Press, Plot No. 2, Phase-IV, Kundli, Haryana.Contents Preface iti Preface to the First Edition av = 1.2 Concept of Data Structures 4 L3__Overview of Data Structures _6 1.4 Implementation of Data Structures 1.5 Organization of the Book 70 Arrays 12-35 - 2 22 Terminology 13 2.3 One-Dimensional Array 14 2.3.1 Memory Allocation for an Array 14 2.3.2 Operations on Arrays 15 2.3.3 Application of Arrays 21 inContents Vii 45 Applications of Stacks. HI 4.5.1 Evaluation of Arithmetic Expressions i 45.2 Code Generation for Stack Machines _121. 4.5.3 Implementation of Recursion 123 5 Fnctorial C: 7 Ds 4.5.5 Quick Sort 128 45.6 Tower of Hanoi Problem 133 4.5.7 Activation Record Management 136 4.6 Problems to Ponder 150 References 152 Queues 153-188 : 34 —Inuoduction 153 Delinits iss 5.3 Representation of Queues 156 5.3.1 Representation of a Queue using an Array 156 5.3.2 Representation of a Queue using a Linked List 159 $4 Various Queue Structures 160 3.4.1 Circular Queue 160 54.2 Deque 164 5.43 Priority Queue 167 5.5 Applications of Queues 172 5.5.1 Simulation __172 5.5.2 CPU Scheduling in_a Multiprogramming Environment 183 5.53 Round Robin Algorithm _/85 $.6__Problems to Ponder 187 References _188 Tables. 189-211 6.1__Rectangular Tables 190 6.2 Jagged Tables 190 6.3 Inverted Tables _ 193 64 Hash Tables __[94 641 Hashing Techniques 194 64.2 Collision Resolution Techniques 199 64.3 Closed Hashing 200 644 Open Hashing 205 645 Comparison of Collision Resolution Techniques 207 6.5 __Problems to Ponder _2/0 References 211viii Contents L_Trees 212-415 7.1 Basic Terminologies 214 7.2 Definition and Concepts 216 7.2.1 Binary Trees 217 7.2.2 Properties of a Binary Tree 2/8 7.3 Representations of Binary Tree 222 Linear Representation of a Binary Tree 223 Linked Representation of a Binary Tree 226 Physical Implementation of a Binary Tree in Memory 228 74 Operations on a Binary Tree 230 14.1___Insertion __230 1.4.2___Deletion _234 1.43 Traversa 232 744 Merging together Two Binary Trees 248 1.5 Types of Binary Trees 249 7.5.1 Expression Tree 250 7.5.2 Binary Search Tree 254 7.5.3____Heap Trees 266 Threaded Binary Trees 276 7.5.5 Height Balanced Binary Tree. 289 15.6 —Red-black Tree __306 15.7 Splay Tree __333 75.8 Weighted Binary Tree 351 759 Decision 7 16 Trees and Forests 366 7.6.1 Representation of Trees 367 L7_B Trees _ 375 1.7.1 __B Tree Indexing 376 172 rations ona B Tree 377 1.17.3 Lower and Upper Bounds of a B Tree 400 78 B+ Tree Indexing 401 19 ___Trie Tree Indexing _ 403 19.1 Trie Structure 404 192 erations on Trie___405 7.9.3 Applications of Tree Indexing 408 Problems to Ponder 4/0 References 414 8. Graphs 416-494 $1 Introduction _46 8.2 Graph Terminologies 418Contents AX. 10. 8.3 Representation of Graphs 422 8.3.1 Set Representation 423 8.3.2 Linked Representation 424 8.3.3. Matrix Representation 425 84 Operations on Graphs 437 84.1 Operations on Linked List Representation of Graphs 437 84. Operations an Matrix Representation of Graphs 444 8.5 Application of Graph Structures 454 8.5.1 Shortest Path Problem _45¢6 8.5.2 Topological Sorting 466 8.5.3 Minimum Spanning Trees 470 8.5.4 Connectivity ina Graph 477 Pr ‘uilicnion Clea 8.6 BDD and Its Applications 487 Convers = A 8 $.6.2 Applications of BDD 490 8.7 Problems to Ponder 492 References 494 Sets. 495-527 9.1 Definition and Terminologies 497 9.2 Representation of Sets 498 9.21 Linked List Representation of Set 498 9.2.2 Hash Table Representation of Sets 499 9.2.3 Bit Vector Representation of Sets 499 924 Tree Representation of Sets 500 9.3 Operations of Sets S506 9.3.1 Operations on List Representation of Set 506 9.3.2 Operations on Hash Table Representation of Set Si 9.3.3 Operations on Bit Vector Representation of Set. $13 934 Operation on Tree Representation of Set S17 9.4 Applications of Sets S2t 941 Spelling Checker J2t 9.4.2 Information System using Bit Swings 522 2s mation Sten oe 9.5 Problems to Ponder $26. References $27 Sorting 528-711 10.1_Basic Terminologies 529 10.2 Sorting Techniques 53Classic Data Structures With the built-in data types, programming languages provide users with a lot of advantages of processing of various types of data, For example, if'a user declares a variable of type Real (say), then several things are automatically implied, such as how to store a value for that variable, what are the different operations possible on that type of data, what amount of memory is required to store, cic, All these things are taken care of by the compiler or the run-time system manager. Abstract data type When an application requires a special kind of data which is not available as a built-in data type, then it is the programmer's responsibility to implement his own kind of data. Here, the programmer has to specify how to store a value for that data, what are the operations that can meaningfully manipulate variables of that kind of data, amount of memory required to store a variable. The programmer has to decide all these things and accordingly implement them. Programmers’ own data type is termed abstract data type. The abstract data type is also alternatively termed user-defined data type. For example, suppose we want to process dates of the form dd/mm/yy. For this, no built-in data type is known in C, FORTRAN, and Pascal. If 4 programmer wants to process dates, then an abstract data type, say Date, has to be devised and various operations such as adding a few days to a date to obtain another date, finding the days between two dates, obtaining a day for a given date, etc, have to be defined accordingly. Besides these, programmers should decide how to store the data, what amount of memory will be needed to represent a value of Date, etc. An abstract data type, in fact, can be built with the help of built-in data types and other abstract data type(s) already built by the programmer. Some programming languages provide a facility to build abstract data types easily. For example, using, struct/class in C/C++, and using record in Paseal, programmers can define their own data types. 1.2 CONCEPT OF DATA STRUCTURES A digital computer can manipulate only primitive data, that is, data in terms of Os and 1s. Manipulation of primitive data is inherent within the computer and does not require any extra effort on the part of the user. But in our real-life applications, various kinds of data other than the primitive data are involved. Manipulation of real-life data (also termed user data) requires the following essential tasks: 1. Storage representation of user data: User data should be stored in such a way that the computer can understand it. 2. Retrieval of stored data: Data stored in a computer should be retrieved in such a way that the user can understand it, 3. Transformation of user data: Various operations which require to be performed on user data so that it can be transformed from one form to another. The basic theory of computer science deals with the manipulation of various kinds of data, wherefrom the concept of data structures comes in. In faet, data structures constitute the fundamentals of computer science. For a given kind of user data, its structure implies the following:Introduction and Overview 5 1. Domain (2): This is the range of values that the data may have. This domain is also termed data object. 2. Function (Z): This is the set of operations which may legally be applied to elements of the data object, This implies that for a data structure, we must specify the set of operations, 3. Axioms (-#): This is the set of rules with which the different operations belonging to 7 can actually be implemented. Now we can define the term data structure. A data structure D is 2 wiplet, that is, D = (D, 9, #) where Dis a set of data object, 7 is aaset of functions and 4 is a set of rules to implement the functions. Let us consider an example. We know that for the integer data type (int) in the C programming language the structure includes the following types: D= (0,41, 32, 23, ..) FHA @ = (A set of binary arithmetics to perform ad multiplication, and modulo operations.) mn, subtraction, division, It can be easily recognized that the triplet (©, 7, #) is nothing but an abstract data type. Also, the elements in set D are not necessarily from primitive data; it may contain elements from some other abstract data types. Alternatively, an implementation of a data structure D is a mapping from ® to a set of other data structures D,, i = 1, 2, .... , for some m. More precisely, this mapping specifies how every object of Dis to be represented by the objects of Dj, = 1, 2, wean. Every function of D must be written using the function of the implementing data structures Dj i= 1, 2, «1. The fact is that each of the implementing data structures is either a primitive data type or an abstract data type. We can conclude the discussion with another example. Suppose, we want to implement a data type, namely Complex as an abstract data type. Any variable of the complex data type has two parts; a real part and an imaginary part. In our usual notation, if z is a complex number, then z = x + iy, where x and y are the real and imaginary parts, respectively. Both x and y are of the Real data type which is another abstract data type (available as a built-in data type). So, the abstract data type Complex can be defined using the data structure Real as Complex z { 4+ Real y : Real ) Now the set D of Complesxc can be realized from the domain of x and y which is Real in this case, Let us specify the set of operations for the Complex data type, which are stated as J F=(@,-,8 +.) Assume that 2, = xy + iy, and 22 = x; + dy ate two data of the Complex data type. Then we can define the rules for implementing the operations in , thus giving axioms #. In the current example, for the Complex data type6 Classic Data Structures a= { 254 82 = y+) + iO + yd (complex addition) 224-22 = — 4p) + i — 2) (complex subtraction) 2% @ Zs (ty XX) — yy X ya) + Hy X yy XX y)) (Complex multiplication) zeata a Na Ya (complex division) aay t ya) ay tye Hj aty? ox ty? relal= yet ty? (complex magnitude) } Note that how different operations of the Complex data type can be implemented using the operations +, -, x, /, of the implementing data structure, namely Real. 2=Vq = (complex conjugate) Assignment 1.1 Implement Dare as an abstract data type which consists of da/mmlyy, where dd varies from 1 to 31, mm varies from 1 to 12 and yy is any integer with four digits. Specify 7 consisting of all possible operations on variables of type Date and then define 4 to implement all the operations in, Assume the implementing data structure(s) which is/are necessary. 13 OVERVIEW OF DATA STRUCTURES In computer science, several data structures are known depending on the areas of application. Out of them, a few data structures are frequently used in almost all application areas and with the help of which almost all complex data structures can be constructed. These data structures are known as fundamental data structures or classic data structures. Figure 1.2 gives a classification of all classic data structures. Classic data structures: Linear Nenlinoar Linear data structures ‘Arrays | [Linked ists] [stacks | [ Queves Nonlinear data evuctures Trees ‘Geephs_] [Tavies Sets Figure 1.2 Classification of classic data structures.4 15 14 1 9 6 7 42 5 10 4 a 16 3 2 13 Figure 2.45 A magic square with SUM = 34, (a) Write a program to read a set of integers for a square matrix and then decide whether the matrix represents a magic square or not. (b) Write a game program as follows: (i) Read the size of the square matrix, N x N. (ii) Display a square matrix (now it is blank) of N x N. Gi) Allow the player to insert data into the matrix as displayed (you should give a chance to the user to confirm the entry and to alter the previous entries, if desired). (iv) After the completion of all the entries from the player, count the score as follows: Score = 0 (zero) if itis not a magic square. Otherwise score = T+ P*100, where T is the time required in seconds and P is the number of alteration of entries, The player's performance will be judged by the minimum score achieved other than zero. REFERENCES Gollieb, C.C. and L.R. Gotlieb, Data Types and Structures, Prentice Hall, Englewood Cliffs, New Jersey, 1986. Horowitz, Ellis, and Sartaj Sahni, Fundamentals of Data Structures, Computer Science Press, Rockville, Maryland, 1985. Kruse, Robert L., Bruce P, Leung and L, Clovis Tondo, Data Structures and Program Design in C, Prentice Hall of India, New Delhi, 1995. Tremblay, Jean Paul, and Paul G. Sorenson, An Introduction to Data Structeres with Applications, McGraw-Hill, New York, 1987.Linked Lists An array is a data structure where elements are stored in consecutive memory locations. In order to occupy the adjacent space, a block of memory that is required for the array should be allocated before hand. Once the memory is allocated, it cannot be extended any more. This is why the array is known as a static data structure. In contrast to this, the linked list is called a dynamic data structure where the amount of memory required can be varied during its use. In the linked list, the adjacency between the elements is maintained by means of links or pointers. A link or pointer actually is the address (memory location) of the subsequent element. Thus, in a linked list, data (actual content) and link (to point to the neat data) both are required to be maintained. An clement in a linked list is a specially termed node, which can be viewed as shown in Figure 3.1. A node consists of two fields: DATA (to store the actual information) and LINK (to point to the next node). LINK |—> Link to the next node Figure 3.1 Node: an element in a linked list.Linked Lists 37 3.1. DEFINITION A linked list is an ordered collection of finite, homogeneous data elements called nodes where the linear order is maintained by means of links or pointers. Depending on the requirements the pointers are maintained, and accordingly the linked list can be classified into three major groups: single linked list, circular linked list, and double linked list 3.2 SINGLE LINKED LIST In a single linked list each node contains only one link which points to the subsequent node in the list. Figure 3.2 shows a linked list with six nodes. Here, NI, N2, ..., N6 are the constituent nodes in the list. HEADER is an empty node (having data content NULL) and only used to store a pointer to the first node NI. Thus, if one knows the address of the HEADER node from the link field of this node, the next node can be traced, and so on, This means that starting from the first node one can reach to the last node whose link field does not contain any address but has a null value. Note that in a single linked list one can move from left to right only; this is why a single linked list is also called one way list. 3.2.1 Representation of a Linked List in Memory There are two ways to represent a linked list in memory: 1. Static representation using array 2. Dynamic representation using free pool of storage Static representation In static representation of a single linked list, two arrays are maintained: one array for data and the other for links. The static representation of the linked list in Figure 3.2 is shown in Figure 3.3. HEADER ¢ 5 NS Figure 3.2 A single linked list with six nodes.38 Classic Data Structures DATA LINK a” 38 42{_— «alta wal _88 45[_ 80 Header 4g} —— al 7 4a 49) 50[ 64 Memory | Array of location pointers, Figure 3.3 Static representation using arrays of the single linked list of Figure 3.20. ‘Two parallel arrays of equal size are allocated which should be sufficient to store the entire linked list. Nevertheless this contradicts the idea of the linked list (that is non-contagious location of elements). But in some programming languages, for example, ALGOL, FORTRAN, BASIC, etc. such a representation is the only representation to manage a linked list. Dynamic representation ‘The efficient way of representing a linked list is using the free pool of storage. In this method, there is a memory bank (which is nothing but a collection of free memory spaces) and a memory manager (a program, in fact). During the creation of a linked list, whenever a node is required the request is placed to the memory manager; the memory manager will then search the memory bank for the block requested and, if found, grants the desired block to the caller. Again, there is also another program called the garbage collector: it plays whenever a node is no more in use; it returns the unused node to the memory bank. It may be noted that memory bank is basically a list of memory spaces which is available to a programmer. Such a memory management is known as dynamic memory management. The dynamic representation of linked list uses the dynamic memory management policy. ‘The mechanism of dynamic representation of single linked list is illustrated in Figures 3.4(a) and 3.4(b). A list of available memory spaces is there whose pointer is stored in AVAIL, For a request of a node, the list AVAIL is searched for the block of right size. If AVAIL is null or if the block of desired size is not found, the memory manager will return a message accordingly, Suppose the block is found and let it be XY. Then the memory manager will return the pointer of XY to the caller in a temporary buffer, say NEW. The newly availed node XY then can be inserted at any position in the linked list by changing the pointers of the concerned nodes. In Figure 3.4(a), the node XY is inserted at the end and change of pointers is shown by the dotted arrows. Figure 3.4(b) explains the mechanism of how a node can be returned from a linked list to the memory bank.Linked Lists 39 Avail + [2] Ey fant} Header m € Non Figure 3.4{a) Allocation of a node from memory bank to a linked list. Availfe} Header yee oes bs Figure 3.4(b) Returning a node from a linked list to memory bank.Linked Lists 84 the free list is saved following an allocation and is used to begin search for the subsequent request. The idea of this strategy is to reduce the search by avoiding examination of smaller blocks that, in the long run, tend to be created at the beginning of the free list as it happens in the case of first-fit allocation. Comparison among the strategies All these strategies are for variable sized blocks. With these strategies, we encounter the problem of fragmentation, a problem which never occurs in the case of fixed size requests. There are two types of memory fragmentation: internal fragmentation and external fragmentation. Internal fragmentation External fragmentation makes the memory management system inefficient as it results in a large umber of small blocks and hence a long list for searching. In order to get rid of this problem, we can refuse to split a block into small pieces, one of which might be very small; instead what we can do is that allocate the entire block which is larger than the requested size. This excess space is termed ‘hole’, which is ‘allocated but unused’. This is wasted in the sense that this hole cannot be allocated any more. This phenomenon of partitioning the total unused storage into available blocks and allocating these blocks with some portion of these blocks remaining unused but not available, is called internal fragmentation. External fragmentation Tf a large number of storage blocks are requested and allocated, the linked list of available storages can be reasonably lengthy and the average block size becomes small and there remain probably very few blocks which are large. If a request for a large block is received, it may have to be denied because there is no single block on the free list that is big enough, even though the total amount of free storage (in the small blocks) may be much greater than the requested amount. This phenomenon of decomposing the total available storage into a large number of relatively small blocks is called external fragmentation. Fragmentation is a major problem in any dynamic memory management system and should be handled carefully, All the above-mentioned strategies suffer from the fragmentation problem. First-fit and best-fit are among the most popular strategies in a dynamic memory management system. As indicated carlier, first-fit is generally faster because it terminates as soon as a free block, large enough to house a new partition, is found. The best-fit method, on the other hand, does not use the first suitable block found, but instead continues the search until the smallest suitable block has been found. This, although, tends to save larger blocks at times, which may be needed later to service large requests, Thus, the best-fit method requires a search of the entire list, while the average length of search for the first-fit would be half of the best- fit or even less. Furthermore, the best-fit method has the unfortunate tendency to produce a large number of very small free blocks, and these are often unusable by almost all requests. In principle, the firstefit is faster, but it does not minimize wastage of memory for a given allocation. Best-fit is slower and it tends to produce small leftover free blocks that may be too small for subsequent allocation. However, when processing a series of requests starting with an initially free memory, neither strategy has been shown to be superior to the other in terms of wasted memory.82 Classic Data Structures Worst-fit prevents what the best-fit does; it reduces the rate of production of small blocks. However, some simulation studies indicate that the worst-fit allocation is not very effective in reducing wasted memory in processing a series of requests. Next-fit as discussed, is a modification of the first-fit strategy. In general, the next-fit does not outperform the first-fit in reducing the amount of wasted memory. So far as processing speed is concerned, a rough comparison can be made as per their order of superiority, which is specified below: Next-fit > First-fit > Best-fit, Worst-fit The boundary tag system uses a slight variation of the above-mentioned strategies. It uses the next-fit storage allocation strategy and does not consider a block for allocation which if allocated leaves a small block of size < €. That is, our pool of free storage will not contain any free block of size < €, where cis chosen by a statistic based on the nature of requests. As per the next-fit strategy, after allocating a block, search for the subsequent request will continue from the next block of these allocated blocks, and let AVAIL store the address of such a next block. Algorithm GetNodeNextFit_ BTS Input: N, the size of the block requested, ¢ the minimum size of a block for fragmentation, and AVAIL being the pointer to the block on the list from where the search for the desired block is to be continued, Ouput: ptr, the pointer of the block of required size if available else message. Data structure: Linked structure for the boundary tag system. Steps: 1, ptr = AVAIL AWAIL is the current pointer to a block in the list 2. flag = 0 11 This flag is used to indicate the continuity of search 3. While (flag = 0) do 4. If (puSIZE > A) then WE large enough block is found 5 diff = ptrsSIZE—N 14 Difference between block size and requirement 6. If (diff < ¢) then Mf Fragmentation test suggests for whole block allocation 1. ptrl = ptr-3LLINK 8 ptr2 = ptr-sRLINK J Delete the block from the list 9. ptrlaRLINK = pir2 10. ptr2—LLINK = ptrl ue pir-sTAG = 1 2. (ptr+ptr SIZE - 1) 3TAG=1 JY Block is allocated 13. AVAIL = pur 11 Block for next search 14, flag = 1 // Block is found 15. Return(ptr) 1 Retum pir to the caller 16. Else HW Allocate the lower words in the block 17. ptrSIZE = diffLinked Lists 83. H Make the upper words in the block free (ptr + diff- 1) =TAG = 0 pwr = pir + diff 11 Allocate the last words of the block and plraSIZE = N If set its fields as allocated ploTAG = 1 (pir+N- 1) 3TAG flag = 1 # Block is allocated Enait Else If (pt 9RLINK = AVAIL) then Print “Block size is insufficient or memory underflow” flag = 1 Exit Else ptr = ptr oRLINK H] Move to the next block Endl Endlf 36. EndWhile 37. Stop It may be noticed that whenever a block is allocated, it is only required to change the field TAG and SIZE; the other fields like LLINK, RLINK and UPLINK are not at all required to be updated, This is because for an allocated block these fields are useless information. 3.9 DEALLOCATION STRATEGY When a block is no more in use, it is required to be retumed to the pool of free storages so that other programs can use this storage space. The boundary tag system not only inserts this block to the list of free blocks, but also combines this newly inserted block with its adjacent block(s). The various cases of this type of combining during the deallocation are illustrated in Figure 3.26. Here, we assume a total 20K memory space and 5 programs, viz. Pl, P2, P3, P4 and PS which are using it, occupying memories of sizes 4K, 2K, 4K, SK, and 2K, respectively. Suppose, P3 releases its space of 4K and it is inserted into the list of free storage. In this sitmation, the available list of storage contains two non-adjacent free blocks of size 4K and 3K, as shown in Figure 3.26(a). Now consider the case when PI releases the block allocated to it. This block has a right adjacent block and hence two blocks will be combined into one. Figure 3.26(b) depicts it. Next, let us consider the case, when P2 releases the block of size 2K. As there are two adjacent free blocks of this block, it is possible to fuse these three blocks into one. Figure 3.26(c) illustrates this case.84 Classic Data Structures 3 freod this block Free spaco P2 Pt 2K Ed Ed 13K 17K 20K 17K ol ole (2) Map for total 20K memory space and its corresponding list of free blocks Ps pa qe K [ae] _ fo] fax) Cx] 1 2K 7K nk 13K 17K 20K (}) Free storage list when 1 releases the block of 4k 2 freed this block a ms aR] el ri ee « 7 ole (0) Free storage space when P2 releases a block of size 2 Figure 3.26 Deallocation of blocks in a boundary tag system. The process of combining two adjacent blocks is very straightforward in the boundary tag system, Here is the use of two TAG fields in each block. Whenever a block is returned, the system will search the TAG fields of its adjacent blocks. If the TAG field(s) is/are found as zero, then combination can be carried out by changing a few pointers. Four cases of combination may arise. Changes in pointers for four cases are illustrated in Figure 3.27.Linked Lists 85 put - Avail Nevily available troo block having no free block{s) adjacent to it ° (a) Case 1: Newly freed block does not have any adjacent blocks on both sides Avail 4 . fol? ° Newly available tree bloek has a laft adjacent {ree block B1 {b) Case 2: Newly freed block has only a left adjacent block as free Figure 3.27 Continued,86 Classic Data Structures pit oie pba x qT x a) ° felol le x x Bt ole a8 ® @ @ Le @Q 7 [elo] _[## Newly available Block 82 has Be t free block Bt 4 Bi ole (€) Case 3: Newy raed block has only aright adjacent bock as foe ptt ptr @ ptrs oe (8) Gase 4; Newly rood block has a left as well as right adjacent free blocks Figure 3.27 Deallocation strategy in a boundary tag system. From Figure 3.27, it can be understood how a very recent feed block can be combined with its adjacent free block(s) to build a larger block in the list of storages. The following is the algorithm ReturnNode_BTS describing the above illustrated cases of combining the free blocks when a free block is returned to the memory system.Linked Lists 87 Algorithm ReturnNode_BTS(PTR) Input: A list of free storage having AVAIL as the pointer to the header of the lists; PTR is the pointer of the block to be deallacated. Output: A list of free storage with newly free block inserted into it. Data structure: Linked link structure for boundary tag system. Steps: 1. n= PTRSSIZE 7* Both adjacent blocks of the newly freed block are not free */ 2. Case 1: ((PTR - 1) TAG = L) and (PTR +) 3TAG = 1) ‘fee Insert at front #4) 3. PTIRSTAG =0 11 Set the TAG fields of the new block as free block 4. (PIR+n-1) 3TAG=0 $. (PTR +2 ~ 1) SUPLINK = PTR 11 Set the UPLINK 6, ptrl = AVAIL>RLINK If Pointer to the first block in the list 7. purlSLLINK = PTR. 11 Change the pointer as shown (1) in Figure 3.27(a) 8. PTRORLINK = purl 11 Change the pointer as shown (2) in Figure 3.27(a) 9. AVAILSRLINK = PTR 1 Change the pointer as shown (3) in Figure 3.27(a) 10. PTRLLINK = AVAIL 11 Change the pointer as shown (4) in Figure 3.27(a) 7* Only left adjacent block to the newly freed block is free *f 11, Case 2: ((PTR ~ 1).TAG = 0) and (PTR + n).TAG = 1) /*** Append after the left adjacent black **7 12. ptrl = (PTR - 1) 9UPLINK #7 Obtain the pointer to the left free block 13, ptrl SIZE = ptrl SIZE +n H Update the size of left block which Hfincludes new free block 14, (pil +1 - 1) STAG =0 7 Set the TAG of the newly combined block 15. (ptrl +n — 1) SUPLINK = ptrl H Set the UPLINK as shown (1) in Figure 3.27(b) /* Only right adjacent block to the newly freed block is free */ 16. Case 3: ((PTR - 1).TAG = 1) and (PTR + n).TAG = 0 /*** Append right adjacent block afier the newly freed block ***/ 18, PTR — 1) UPLINK 19, ptrl>RLINK 20. ptr3 = ptr2-RLINK 21. pt3LLINK = pirl 1 Change of pointer as shown (1) in Figure 3.27(c) 22. PTRORLINK = ptr3 M# Change of pointer as shown (2) in Figure 3.27(c) 23. PTRLLINK = pirl M Change of pointer as shown (3) in Figure 3.27(c) 24. ptrl-+RLINK = PTR M Change of pointer as shown (4) in Figure 3.27(c) 28. PTRSIZE = n + ptr2eSIZE H Change the size of the newly combined block 26, (PTR +n - 1) SUPLINK = PTR H Set UPLINK fields of the newly 4 combined block 27. (pirl +n - 1) STAG = 0 iI Set TAG field of the newly combined block /* Both the left and right adjacent blocks to the newly freed block are free %/ 28. Case 4: ((PTR - 1) TAG = 0) and ((PTR +n) STAG = 0) Pe Append new free block after left adjacent block and right adjacent block **/ (Contd)88 Classic Data Structures (PTR — 1) 2UPLINK if Pointer to the left adjacent block W Pointer to the right adjacent block 31, pt3 = pr? RLINK 32, ptrl SRLINK 33, pt39LLINK # Change of pointer as shown (1) in Figure 3.27(d) # Change of pointer as shown (2) in Figure 3.27(d) 34, trl SIZE = pirl SIZE + n + pir20SIZE It Change the size of newly 1 combined block 35. (ptrl + ptrlSIZE ~ 1) STAG = 0 11 Set the TAG field of the i newly combined block 36. (ptrl + ptrlsSIZE- 1) UPLINK = piel i Set UPLINK field as shown (3) fin Fig. 3.27(€) 37. Stop Assignment 3.9 (a) What modification would you suggest in order to make the algorithm GetNodeNextFit_BTS as GetNodeFirstFit_BTS based on the first-fit allocation strategy? (6) Write an algorithm for availing a node based on the best-fit storage allocation strategy. (©) Procedure ReturnNode_BTS maintains a list of free storages. For a given request, this list is to be searched from header node whose pointer is AVAIL. One modification is suggested as “Recently freed block is for the consideration of next service”. What necessary change you should incorporate in the procedure ReturnNode_BTS and | GetNodeNextFit_BTS to incorporate the suggested modification? 3.10 BUDDY SYSTEM So far we have discussed that the block sizes are either fixed or completely arbitrary. The buddy system is another storage management system which restricts the sizes of blocks to some fixed set of sizes, These blocks of restricted sizes are maintained in a linked list. Whenever a request for a block of size N comes, the number M, the smallest of the fixed sizes but equal to or larger than N, is determined, and a block of size M is allocated, if available on the list. If a block of size M is not available, then a larger block, if available, is split into two sub-blocks (known as buddies), each of them are also of fixed sizes, and this process is repeated until a black of size M is produced. ‘A buddy system specifies the restricted sizes as Fo. Fj, ... Fax for blocks according to some pattern, One principle for the specification of block sizes is the use of the recurrence relation Fy = Frit Fre kSnSMAX for a given k and MAX Fy = Cy Fy = Cy, mn Fe = Ce with initial conditions.Linked Lists 89 For example, if k = 1 and Fy = 8, then the block sizes are 8, 16, 32, 64, 128, .... That is, the block sizes are successive powers of 2; and the buddy system based on such fixed sizes is called the binary buddy system. Another buddy system which restricts the fixed sizes to the Fibonacci sequence is called the Fibonacci buddy system. This system is for k= 2 and Fy = 8, F, = 13; the sizes of the blocks then are 8, 13, 21, 34, 55, 89, ..., etc. Note that, Fo, Fy, Fz, etc. ‘are specified based on the applications, that is, how much small sizes of the blocks it may require, ‘When a request for a memory space comes to the system, it searches for a block whose size is nearer to but not smaller than the size of the requested block; if such a block is found then it will be allocated; otherwise the next larger block or the largest block in the list (depending ‘on the allocation strategy) will be split successively till the block having the size nearest to the tequired size is found. On the other hand, whenever a block is returned to the system, it will ‘be combined with another free block into a larger free block if these two blocks were the ‘buddies that were formed by some previous splitting. If this larger free block is again qualified as a buddy, then it will be recombined with its partner buddy, if it is free and this process of teformation will continue till no buddies fit for coalescing are present. To illustrate the above allocation and deallocation techniques, let us consider the Fibonacci buddy system which restricts the block sizes as 8, 13, 21, 34, 55, 89 and 1. sume that 144 is the largest block possible, that is, maximum memory available. Figure 3.28 illustrates a few cases of allocation and deallocation. From Figure 3.28, the following facts are evident: initially the system maintains a free block of size 144, say. When a request for a block of size 30 comes, it splits into two buddies, namely 89 and 55 and SS is again split into two buddies, namely 34 and 21. As 34 is the nearest block size so it is allocated. Figure 3.28(a) illustrates this. Now for a subsequent request for a block of size 50, it searches and finds that a block of size 89 is available. As 89 is too large for the request, a split occurs into 55 and 34 and then the block of size 55 is allocated for the request. This is illustrated by Figure 3.28(b). Figure 3,28(c) illustrates the case when a block of size 55 is returned to the system. As 55 and 34 are free buddies, so they are combined into a block of size 89, Block 89 has no free buddy, so here merging comes to a halt. On the other hand, deallocation for a block of size 34 (Figure 3.28(d)] first combines 34 and 21 as they are free buddies into a large block of size 55. Next, 55 and 89 are free buddies, so another combination yields the largest block of size 144. ‘Our next discussion is to define the structure of a node which will be suitable to manage the above-mentioned operations. For this we can decide the following structure: LuNK+—fe] FREE | SiZE [CODE] #}—- RLINK Two link fields, LLINK and RLINK, store the addresses of the predecessor and the successor of the block, The SIZE field stores the size of the block. Instead of storing the absolute size of the block we will store the index value for the size in the recurrence relation which the buddy system assumes. For example, in the case of Fibonacci buddy system, F, = Fy + Fray With Fo = 8, Fy = 13, we will assume the value for SIZE field as90 Classic Data Structures Absolute size (F;) SIZE (n) Fy=8 0 F,=13 1 Fy=21 2 Fy = 34 3 Fy = 55 4 Fs = 89 5 Fe = 14d 6 and so on, Another field CODE has enough significance in this system, This field will store the information whether two blocks are buddies or not, A clever solution has been suggested by Hinds (in Proceedings of IEEE Symposium on Foundations of Computer Science, Vol, 19, 123— 130, 1978) in order to decide the buddy property of blocks in a Buddy system; the solution is mentioned below: Initially: CODE =0 IThis is for the largest block in the system Splitting: CODE,eer = CODEparext + 1 On splitting a parent block, two blocks CODEggut Mnamely LEFT and RIGHT will result Merging: CODEparenr = CODE,uer-1 —//On merging of two buddies the resultant Molock has CODE which is one less than its Meft buddy a NN ———$§{$§ 14 —____- [Tt 21 99 84 {a) Allocation for a request of block size 30 144, a < ~ =u 34 < 4 > —s5— ar {0) Anather allocation in response to a request for a block of size 50 144 (6s) A 1 @) 3 a 9 tri (€) Deattocation of block size 55 Figure 3.28 Continued.102 Classic Data Structures 32 33 a4 35 3.6 37 Swap two adjacent elements (a) by interchanging the elements (b) by adjusting only the pointers (and not the data). In both cases, assume that the elements are stored in (i) a single linked list (ii) a double linked list. Which method out of (i) and (ii) might appear better? Why? Given a single circular Linked list containing a set of data, obtain the following from this data structure: (a) Reverse the direction of links. (b) For the given two elements in the list, find the distance (that is, the number of nodes) between them. Write an algorithm and implement it with C++ in each case. Given a single linked list containing any type of data, obtain the following: (a) Reverse the ordering of the data. (b) Suppose X and ¥ are two nodes in the list. Add all the nodes between X and Y (both are inclusive) so that X is the first node in the list. (ii) is the tast node in the list. Let A and B be two lists representing two polynomials. Obtain the operation to find (a) C =A + B, where C is a list representing a polynomial and is obtained by adding two polynomials A and B. (b) C=A-— B, where Cis a list representing a polynomial and is obtained by subtracting two polynomials A and B. (©) C =A*B, where C is a list representing a polynomial and is obtained by multiplicating two polynomials A and B. The following are the sorting algorithms which can be performed on the set of data which are stored in a single linked list. (a) Bubble sort (b) Selection sort (©) Insertion sort (@) Radix sort (e) Quick sort Write the genetic code for the above-mentioned sorting methods using the linked list data structure, Obtain a comparative study of the above sorting methods. Suppose two lists INPUT and SPLITTER containing data of the same type are given, as shown in Figure 3.34. All the elements are to be split into three lists, LIST1, LIST2 and LIST3, based on the SPLITTER so that LIST! contains all the data less than any data in SPLITTER, LIST2 contains data equal to any data in SPLITTER and LIST3 contains all the data greater than any data in SPLITTER. Write an algorithm for this. Implement the algorithm with C++.Linked Lists 103 Rem htop [nema u emt Size=ut/—1 — {a) Array representation of a stack (0) Linked list representation of a stack Figure 4.3 Two ways of representing stacks. 43.2 Linked List Representation of Stacks Although array representation of stacks is very easy and convenient but it allows the representation of only fixed sized stacks, In several applications, the size of the stack may vary during program execution. An obvious solution to this problem is to represent a stack using @ linked list. A single linked list structure is sufficient to represent any stack. Here, the DATA. field is for the ITEM, and the LINK field is, as usual, to point to the next item. Figure 4.3(b) depicts such a stack using a single linked list. In the linked list representation, the first nede on the list is the current item that is the item at the top of the stack and the last node is the node containing the bottom-most item. Thus, a PUSH operation will add a new node in the front and a POP operation will remove a node from the front of the list. The SIZE of the stack is not important here because this representation allows dynamic stacks instead of static stacks, as with arrays.108 Classic Data Structures In the linked list representation of a stack, whether a stack is empty or not can be ascertained by testing the LINK field of the STACK_HEAD node. Note that a test for overflow is not applicable in this case. 4.4 OPERATIONS ON STACKS The basic operations required to manipulate a stack are: PUSH To insert an item into a stack POP To remove an item from a stack STATUS To know the present state of a stack Let us define all these operations of a stack. First, we will consider the above-mentioned operations for a stack represented by an array. Algorithm Push_Array Input: The new item ITEM to be pushed onto it. Ouiput: A stack with a newly pushed ITEM at the TOP position. Data structure: An array A with TOP as the pointer. Steps: 1, If TOP 2 SIZE then 2. Print “Stack is full” 3. Else 4. TOP = TOP +1 8, A[TOP] = ITEM 6. Endlf 7. Stop Here, we have assumed that the array index varies from 1 to SIZE and TOP points the location of the current top-most item in the stack. The following algorithm Pop_Array defines the POP of an item from a stack which is represented using an array A. Algorithm Pop_Array Input: A stack with elements. Output: Removes an ITEM from the top of the stack if it is not empty. Data structure: An array A with TOP as the pointer. Steps: 1, If TOP <1 then Print “Stack is empty” Eke ITEM = A[TOP] TOP = TOP -1 Endlf 7. Stop aveepStacks 109 In the following algorithm Stams_Array, we test the various states of a stack such as whether it is full or empty, how many items are right now in it, and read the current element at the top without removing it, etc. Algorithm Status_Array Input: A stack with clements. Owiput: States whether it is empty or full, available free space and item at TOP. Data structure: An array A with TOP as the pointer. Steps: If TOP < 1 then Print “Stack is empty” Else If (TOP 2 SIZE) then Print “Stack is full” Else Print “The element at TOP is”, A[TOP] free = (SIZE — TOPYSIZE * 100 Print “Percentage of free stack is”, free Endlf Endlf Now let us see how the same operations can be defined for a stack represented with a single linked list. Algorithm Push_LL Input; ITEM is the item to be inserted. Oviput: A single linked list with a newly inserted node with data content ITEM. Data structure: A single linked list structure whose pointer to the header is known from STACK_HEAD and TOP is the pointer to the first node. ‘Steps: 1. new = GetNode(NODE) /* Insert at front 4 new-DATA = ITEM . new—LINK = TOP TOP = new STACK_HEAD-LINK = TOP Stop Algorithm Pop_LL Input: A stack with elements, Output: The removed item is stored in ITEM. Data structure: A single linked list structure whose pointer to the header is known from STACK_HEAD and TOP is the pointer to the first node. awayn110 Classic Data Structures Steps: . 1. If TOP = NULL 2. Print “Stack is empty” 3. Exit 4. Else 5. ptr = TOP+LINK 6. ITEM = TOP»DATA 7. STACK_HEADSLINK = ptr 8 TOP =pir 9. Endlf 0. Stop The operation Stars now can be defined as follows: Algorithm Status_LL() input: A stack with elements. Output: Status information such as its state (empty or full), number of items, item at the TOP. Data structure: A single linked list structure whose pointer to the header is known from STACK_HEAD and TOP is the pointer to the first node, ‘Steps: 1, ptr = STACK_HEADLINK 2, Mf (ptr = NULL) then 3. Print “Stack is empty" 4, Else S$. — nodeCount = 0 6 While (ptr # NULL) do 7. nodeCount = nodeCount + I 8 ptr = ptr>LINK. 9. EndWhile 0. Print “The item at the front is”, TOP-DATA, “Stack contains”, nodeCount, “Number of items” 11. Endif 12. Stop Assignment 4.1 Using only PUSH, POP and STATUS as the basic operations defined in Section 4.4, write the procedures to solve the following problems. (a) Read the top-most element in the stack. (Note that read will not remove the clement.) (b) Visit all elements in a stack. (Preferably without using another stack). (©) Search for an item in the stack. (Note: You should consider both types of memory representations of the stack.)Stacks 111. Assignment 4.2 (Multiple stack) In several applications, more than one stack may be required together. Some stacks overflow whereas others are nearly empty. Suppose an application requires two stacks X and Y (Figure 4.4). One can define an array A with N, elements for stack X and another array B with N, elements for stack ¥. Now instead of defining two separate arrays A and B, we can define a single array, say AB, with N = N, + N, clements for X and ¥ together. Let us define the starting locations of items for stack X and stack Y as AB(1] and ABIN] respectively and X grows’ to the right whereas ¥ ‘grows’ to the left. }-*TOPX TOPY<-+| ‘Stack X Stack ¥ Figure 4.4 A multiple stack. With this scheme, overflow will occur only when X and ¥ together have more than N elements. This technique will usually decrease the number of situations of occurrence of overflow even though we have not increased the total amount of space reserved for the two stacks. Using this scheme, we need the modified versions of PUSH_A and POP_A operations as PUSH_X, PUSH_Y, POP_X, and POP_Y. Similarly, the operation STATUS_AB has to be defined to test the state of empty or full, percentage of space occupied by X and Y, etc. Write the basic operations for the implementation of such a multi-stack. 4.5 APPLICATIONS OF STACKS ‘Various applications of stacks are known. A classical application in a compiler design is the evaluation of arithmetic expressions; here the compiler uses a stack to translate an input arithmetic expression into its corresponding object code. Some machines are also known which use built-in stack hardware called ‘stack machine’. Another important application of a stack is during the execution of recursive programs; some programming languages use stacks to run recursive programs, One important feature of any programming language is the binding of memory variables. Such binding is determined by the scope rules, There are two scope rules Known: the static scope rule and the dynamic scope rule, Implementation of such scope rules is possible using a stack known as a run time stack. The following subsections highlight the above-mentioned applications of stacks, 45.1 Evaluation of Arithmetic Expressions An arithmetic expression consists of operands and operators. Operands are variables or constants and operators are of various types such as arithmetic unary and binary operators [for example, - (unary), # (addition), — (subtraction), * (multiplication), / (division), * (exponentiation), % (remainder modulo), etc.], relational operators (for example, <, >, <=,112 Classic Data Structures <>, >=, ele.), and Boolean operators (such as, AND, OR, NOT, XOR, etc.). In addition to these, parentheses such as ‘(’ and ‘)’ ate also used. A simple arithmetic expression is cited below: A+B*C/D-E*F*G The problem to evaluate this expression is the order of evaluation. There are two ways to fix it. First, we can assign to each operator a precedence and associativity. For example, a set of usual operators with their precedence and associativity is given in Table 4.1. Table 4.1 Precedence and associativity of operators Operators Precedence Associativity = (unary), +(unary), NOT 6 - * (exponentiation) 6 Right to left * (multiplication), / (division) 5 Left to right + (addition), — (subtraction) 4 Left to right << <2 2s 3 Left to right AND 2 Left to right OR, XOR 1 Left to right Thus, with the above rules of precedence and associativity of operators, the evaluation will take place for the above-mentioned expression in the sequence (sequence is according to the number 1, 2, 3, ..., etc.) stated below: It should be noted that the above rules for precedence and associativity vary from one programming language to another. ‘Another way of fixing the order of evaluation is parenthesizing the expression fully; this allows one to override the rules for precedence and associativity. The following is the parenthesized version of the same expression: Input: ((A + B) * ((C/D) - (E 4 F * G)) (A fully parenthesized expression) With this parenthesization, the innermost parenthesis part (called sub expression) will be evaluated first, then the next innermost, and so on; such a sequence of evaluations is shown below:Stacks 113, ((A+B)* ((G/D)-(EO(F *G)) VV 1 2 y Whatever way we may specify the order of evaluations, the problem is that we must scan the expression from left to right repeatedly. Hence, the above-mentioned processes are inefficient because of the repeated scanning required. Another problem is the ambiguity about how the compiler can resolve to generate a correct code for a given expression. The last problem mainly occurs for a partially parenthesized expression. These problems can be solved in the following two steps: 1, Conversion of a given expression into a special notation 2. Evaluation/production of an object code using a stack. 3 é Notations for arithmetic expressions There are three notations to represent an arithmetic expression, viz. infix, prefix and postfix (or suffix). The conventional way of writing an expression is called infix. For example, A+B, C-D, E*F, GI/H, ete. Here, the notation is . This is called infix because the operator comes in between the operands. The prefix notation, ‘on the other hand, uses the convention Here, the operator come before the operands. The following are simple expressions in prefix notation: +AB, -CD, *EF, /GH, etc. The prefix notation was introduced by the Polish mathematician Jan Lukasiewicz and hence also termed Polish notation. The last notation is called the postfix (or suffix) notation where the operator is suffixed by operands: The following expressions are in postfix notation: AB+, CD-, EF*, GH/, etc. The postfix notation is just reverse of the Polish notation, hence it is also termed reversed Polish notation.114 _ Classic Data Structures Tt may be noted that in all of the above notations, a unary operator precedes its operand, ‘An expression given in infix notation can easily be converted ‘into its equivalent prefix or postfix notation, The following rule is applied to convert an infix expression into a postfix form. ‘© Assume the fully parenthesized version of the infix expression. ® Move all operators so that they replace their corresponding right part of parentheses. Remove all parentheses. The following example illustrates this conversion. For simplicity, let us consider a fully parenthesized expression. Input: (A + ((B * €) ~ D)) * (E - (A/C) (A fully parenthesized expression) (As((BAC)-D))*(E-(A/C))) UU | ut (Arrows point from operators to their corresponding right parenthesis.) (A (BC *D-+(E (AC T-* (Operators are moved to their respective right parentheses.) Output: ABC*D—-+EAC/—* (AIL parentheses are removed yielding the postfix expression.) A similar technique can be applied to obtain the prefix notation for a given infix notation but moving the operators corresponds to the left parenthesis. Three notations for the given arithmetic expression are listed below: Infix: (A + (B * C) - D)) * (E = (AIC) Prefix: * + A - “ BCD - E/AC Postfix: ABC © D- + EAC /~ * The following points may be observed from the above three notations: 1. In both prefix and postfix equivalents of an infix expression, the variables are in the same relative positions, 2. The expressions in prefix or postfix form are completely parenthesis free. 3. The operators are rearranged according to the rules of precedence of operators. Out of these three notations, the postfix notation has certain advantages over the other notations from the computational point of view. The main advantage of postfix is its evaluation, During the evaluation of an expression in postfix notation it is no longer required to scan the expression from left to right several times, but scanning is required exactly once. This is possible using a stack and will be discussed shortly.Stacks 115 Thus, evaluation of an expression is a two-step process. First, we have to convert the expression into its postfix notation, and then evaluate this expression in postfix notation. In each step, the stack is the main data structure that is used to accomplish these tasks. The uses of the stack for the purpose and the above-mentioned procedures are discussed below. We will assume the array representation of a stack in our discussions. Conversion of an Infix expression to postfix expression To formalize the conversion method, we will assume simple arithmetic expressions containing the +, -, *, /, and “ (exponentiation) operators only (i.e. without unary operators, Boolean operators and relational operators). The expression may be parenthesized or unparenthesized, First, we have to append the symbol *)' as the delimiter at the end of a given infix expression and initialize the stack with ‘(’. These symbols ensure that either the input or the stack is exhausted. ‘Our next step is iterative: read one input symbol at a time and decide whether it has to be pushed onto the stack or not, This decision will be govemed by Table 4.2. Table 4.2. In-stack and in-coming priorities of symbols In-stack Incoming priority value priority value +- 2 1 4 3 a 5 6 operand 8 7 ¢ 0 9 ) - 0 From the table, it can be noted that for a symbol we have considered two priority values, viz. in-stack priority and in-coming priority values. A symbol will be pushed onto the stack if -coming priority value is greater than the in-stack priority value of the top-most element, Similarly, a symbol will be popped from the stack if its in-stack priority value is greater than or equal to the in-coming priority value of the in-coming element. In order to define the algorithm, we will assume the following functions: ReadSymbol(): From a given infix expression, this will read the next symbol. ISP(X): Returns the in-stack priority value for a symbol X. ICP(X): This function retums the in-coming priority value for a symbol X. Output(x): Append the symbol X into the resultant expression. Let as assume that a stack of capacity SIZE is known and TOP is the current pointer in it. PUSH and POP are usual operations of the stack. Algorithm InfixToPostfix input: E, simple arithmetic expression in infix notation delimited at the end by the right parenthesis ‘)’, incoming and in-stack priority values for all possible symbols in an arithmetic expression.116 Classic Data Structures Owput: An arithmetic expression in postfix notation. Data structure: Array representation of a stack with TOP as the pointer to the top-most element. Steps: 1. TOP = 0, PUSH(‘(*) 11 Initialize the stack 2. While (TOP > 0) do 3. item = E,ReadSymbol() W Scan the next symbol in infix expression 4. x= POP) 1 Get the next item from the stack 5. Case: item = operand 17 the symbol is an operand 6 PUSH(x) I The stack will remain same 1. Output(item) 11 Add the symbol into the output expression 8. Case: item = *). J Scan reaches to its end 9. While x #'C do Till the left match is not found 10. Output(x) i x= POP() 12 EndWhile 13, Case: ESP(x) 2 ICP(item) 4. While (ISP(x) > ICP(item)) do 15. Output(x) 16. x= POP() 1 EndWhile 18. PUSH(Q) 19. PUSH (item) 20. Cases ESP(x) < ICP(item) 2. PUSH(x) 22, PUSH (item) 23. Otherwise: Print “Invalid expression” 24, EndWhile 25. Stop Note: This is a procedure irrespective of the type of memory representation of the stack to convert an infix expression to its postfix form using the basic operations of the stack, namely PUSH and POP. These operations can be replaced with their respective versions and hence implementable to stack with any type of memory representation. EXAMPLE 4.1 Let us illustrate the procedure InfixFoPostfix with the following arithmetic expression: Inputs (A+B) * C ~ (D*E)/F) (infix form) Symboi reading: 123.45 67 8 91011 12131415 16Stacks 117 Stack Ourput ( « « A G A G AB c ABt « ABt ia AB+C c ABCA C( ABHCA (-¢ AB+CAD Co AB+CAD (-(* AB +C* DE (ie AB + C*DE* (-/ AB+C*DE* cl AB+CADESF AB+CADE*F Output: A B+ C* DE * F/— (postfix form) The above procedure assumes that the input infix expressions are according to the right syntax So, if the input expression is not correct, its postfix form will not be correct. Extending the same idea, we can incorporate relational, Boolean and unary operators in the above procedure. Assignment 4.3 (a) Verify the algorithm InfixToPosifix for producing the postfix notations of the following from the infix notations. @) (A+ B)*C*D*E-F/G (i) A+ *B=C (iii) (A + B) * (C= D) (iv) A * @B)-C/D (b) Modify the algorithm MnfixToPostfix to convert an infix expression containing Boolean ‘operators and relational operators and unary operators to their equivalent postfix form. (c) For the algorithm InfixToPosifix, we assume that the input expression is correct. That is, there is no error checking performed. But the input expression may not be well formed always: expressions having two operands together or two binary operators together or unmatched parentheses are some common mistakes, Write a procedure, so that it will check for the correct expression at the time of producing the postfix form, Evaluation of a postfix expression For a given expression in postfix notation, it can be easily evaluated. The following algorithm EvaluatePosifix is used to evaluate an arithmetic expression in postfix notation using a stack.118 Classic Data Structures Algorithm EvaluatePostfix Input: E, an expression in postfix notation, with values of the operands appearing in the expression. Output: Value of the expression. Data structure: Array representation of a stack with TOP as the pointer to the top-most element. Steps: 1. Append a special delimiter ‘#° at the end of the expression 2. item = E.ReadSymbol() 1! Read the first symbol from E 3. While (item # ‘#') do 4. If (item = operand) then 5. PUSH(item) 11 Operand is the first push into the stack 6. 1. 11 The item is an operator 8. y = POPC) 11 The right-most operand of the current operator 9. x= POP: M4 The left-most operand of the current operator 10. b=xopy 1 Besform the operation with operator ‘op’ and operands x, LL. PUSH(:) J} Push the result into stack 12, Endlf 13. item = E.ReadSymbol() #! Read the next item from E 14, End While 15, value = POPC) H Get the value of the expression 16. Return(valuc) 17. Stop EXAMPLE 4.2 To illustrate the algorithm EvaluatePostfix, let us consider the following expression: Infix: A+ (B* C)/D Postfix: ABC *#D/+ Input: AB C*D/+# with A= 2, B = 3, C = 4, and D =6 Read symbol Stack A 2 PUSH(A = 2) B 23 PUSH® = 3) Cc 234 PUSH(C = 4) * 212 POP(4), POP(3), PUSH(T = 12) D 2126 PUSH(D = 6) é 22 POP(6), POP(12), PUSH(T = 2) + 4 POP(2), POP(2), PUSH(T = 4) * value = POP()Stacks 119 Conversion of a postfix expression ta a code Using a stack, we can easily generate an assembly code for an expression given in reverse Polish notation (postfix). In order to simplify our example, we will assume the arithmetic expressions with four arithmetic operations— + (addition), - (subtraction), * (multiplication) and / (division) only—and the assembly codes are in single address form. The following assembly code mnemonics are assumed: LDA A To load the secumulator with the memory content of A and the content of A remain unchanged. STAB To store the content of the accumulator in memory location B. ADD A To add the value of memory content A with the value of the accumulator and the result will be stored in the accumulator; the value of memory content A will remain unchanged. SUB B To subtract the value of memory content B from the value of the accumulator and the result will be stored in the accumulator; the memory content B will remain unchanged, MUL C To multiply the value of memory content C with the value of the accumulator and the result will be stored in the accumulator; the memory content C will remain unchanged. DIV D To divide the value of the accumulator by the value of the memory content D and the result of the division will be stored in the accumulator; the value of the memory content D will remain unchanged. EXAMPLE 4.3 For example, let the infix expression be A + B and its postfix form be AB+. The assembly code for this expression will be LDA A ADD B STAT For writing such codes, let us assume one procedure ProduceCode(A, B, op, Temp) with four arguments, For instance, with AB +, op is ADD and Temp is T. With these, the algorithm for converting a postfix expression to ils equivalent assembly code, PosifixToCode, is framed as follows: Algorithm PostfixToCode An arithmetic expression E in postfix notation. 2 Assembly code, Data structure: A stack with TOP as the pointer to the top-most element.120 Classic Data Structures Steps: 1. Append a delimiter ‘#” at the and of the expression 2. item = E.ReadSymbol() If Read the first symbol from the expression 3. f= 1, TOP=0 # Stack is initialized; an integer will be used as index 4. While (item 4 ‘#) do 5. Cases item = operand 6. PUSH(tem) #7 Push the item into the stack 7. Case: item = 4" 8. x= POP() Hf Pop two operands from the stack 9. y = POP() 10. ProduceCodety, x. ‘ADD’, Ti) 11 Ti is the ith temporary M. PUSH (Ti) 12. Cases item =~" 13. x = POP() MW y= POPR() 15. ProduceCodety, x, ‘SUB’, 71) 16. PUSH(Ti) 17. Case: item =“ 18. x= POP() 19. y = POP() 20. ProduceCode(y, x, ‘MUL', 7) 21, PUSH(Ti) 22. Cases item = ‘/ 23. POP() m4. y= POP() 25. ProduceCode (y. x, ‘DIV', Ti) 26. PUSH!) 27. Otherwise: 28. Print “Error in input” 29, Exit 30. item = E.ReadSymbol() W Read for the next symbol from E 31 feat 1 The index is incremented 32. EndWhile 33. Stop EXAMPLE 4.4 The above algorithm is illustrated with the following example: Infix: (A +B) * C/D Postfix: AB+C*D/ Input: AB+C *#DI#Stacks 121 The production of codes according to the algorithm PosifixToCode is given below: Scanned Content of stack ‘Action Code generated symbol A A PUSH A) B AB PUSH(B) + TI x=By=A LDAA PRODUCE_CODE(A, B, ‘ADD’,T1) ADD B PUSH(T1) STATI ic T1C PUSH(C) . T2 x=Cy=Tl LDA Ti PRODUCE_CODE(TI, C, ‘MUL’,T2) MUL C PUSH(T2) STA T2 D T2D PUSH(D) t 73 x=D,y=T2 LDA T2 PRODUCE_CODE(T2, D, ‘DIV’, T3) DIV D PUSH(T3) STA T3 # 13 Stop Assignment 4.4 (a) Consider an expression which contains relational and Boolean operators. Formulate the precedence values required for these operators to convert such an expression to reversed Polish notation. (b) Modify the algorithm PosifoxToCode so that it will also handle the unary operators. (c) Modify the algorithm PostfoxToCode for the six relational operators <, >, <=, >=, <>, =) and four Boolean operators (AND, OR, NOT, XOR). 4.5.2 Code Generation for Stack Machines In Section 4.5.1, we discussed the generation of codes for a given arithmetic expression in postfix form. The codes are of the type called single address codes. These codes are for those machines which maintain a number of registers; the registers are to store the temporaries. One problem using other machines, which have a very limited number of registers, is how to handle the storage of intermediate results. Some machine architectures are known which use a stack instead of registers to store temporaries or intermediate: results; obviously the stack size in this case should be adequate enough to handle a large expression. Here, we will present the description of a simple hypothetical machine to illustrate the concept of code generation for the postfix form of arithmetic expressions. Let us assume the following set of instructions (given in mnemonic forms) of the stack machine.122 Classic Data Structures PUSH To load from memory onto stack, This instruction loads an operand from the memory location named and places the content of into the stack. POP To store the top element of the stack in memory. The content of the top of the stack is removed and stored in the memory location referenced by . Let us assume that our machine performs the following arithmetic operations only: ADD, SUB, MUL, DIV ‘Assume that the operations can be stated with zero-address operation code. Operands are explicitly mentioned as top two elements in the stack. The operations are as stated below: S{TOP - 1] = S[TOP - 1] 8 S[TOP] TOP = TOP - 1 where 0 stands for an operation. This simply means that, for an operation, the operands are from the top two elements, the result is stored at the second top location and the stack pointer is decremented by one. To illustrate the hypothetical machine, let us consider the following arithmetic expressions: A=B*C-A The corresponding postfix notation can be obtained as follows: ABC*A-== The instruction code according to the stack machine is as given below: PUSH A # Load operand A into the stack PUSH B # Load operand B into the stack PUSH C H Load operand C into the stack MUL 1! Muhiply B * C PUSH A Hf Load operand A into the stack SUB # Subwact B* C-A POP A #f Store the result in the memory location for A The various states of the stack are depicted in Figure 4.5. A TK A A A A Empty PushA = PushB = PushG = MUL PushA = SUB. POP Figure 4.5 Evaluation of an arithmetic expression using a stack machine.Stacks 123 To generate machine codes for a stack machine when an arithmetic expression is given in postfix notation, an algorithm Post/ixToCodeForStackMachine is described below. Algorithm PostfixToCodeForStackMachine Input: An arithmetic expression £ in postfix notation. Output: Equivalent codes for stack machine Data structure: Array representation of a stack with TOP as the pointer to the top-most element. Steps: 1, Add a delimiter ‘#' at the end of the expression 2. item = E.ReadSymbol() 4 To read an element from the expression E 3. While (item # ‘#") do 4. If (item = anOperand) i For the operand only 5. ProduceCode(*PUSH’, item) 6. Else W Wtem is the operator 1, Case: item = + 8. ProduceCode (‘ADD') 9. Case: item = 10. ProduceCode (‘SUB") u Case: item = “*" 12. ProduceCode (‘MUL’) 13. Case: item = */" 14, ProduceCode (‘DIV’) Is. Endl 16. item = E.ReadSymbol() 41 Read the next symbol from E 17. EndWhile 18, Stop Here, we assume the procedure ProdwceCode() which will add the code into the list of codes. Tt may be observed that the above-mentioned procedure PostfixFoCodeForStackMachine generates the optimized code. Here, the stack operations PUSH and POP are only involved, which perfectly fit the process of generating the object code from the postfix notation. 45.3 Implementation of Recursion Recursion is an important tool to describe a procedure having several repetitions of the same. A procedure is termed recursive if the procedure is defined by itself. As a simple example, let us consider the case of calculation of the factorial value for an integer n. misnx(n=1)xX (n=2)x"x3x2x1 or nl =n x (n- It ‘The last expression is the recursive description of the factorial whereas the first is the iterative definition, Using a pseudo code, the above two types of definitions are expressed as follows:124 Classic Data Structures Factorial_I Input: An integer number N. Output: The factoral valuc of N, that is Nt Remark: Code using the iterative definition of factorial. . Return(fact) 47 Return the result Stop Here, Step 2 defines the iterative definition for the calculation of a factorial. Now, let us see the recursive definition of the same. Factorial_R Code using the recursive definition of factorial Input: An integer number N. Output: The factoral value of N, that is N!. Remark: Code using the recursive definition of factorial. Steps: 1. If (= 0) then 1 Termination condition of repetition 2. fact =1 3. Else 4. fact = N * Factorial R(W ~ 1) 5. EndIf 6. Return(fact) J Return the result 7. Stop Here, Step 4 recursively defines the factorial of an integer NV. This is actually a direct translation of nl =n * (n = 1)! in the form of Factorial_R (n) = n * Factorial_R (n = 1). This definition implies that n! will be calculated if (1 ~ 1)! is known, which in turn if (n = 2)! is known and so on until 2 = 0 when it returns 1 (Step 1). The question that arises is how this definition can be implemented. We will see that a data structure stack can be used for this purpose. Some programming Innguages such as C, Pascal, which have a dynamic memory management mechanism, can directly accept the recursive definition of procedures: compilers of these programming languages are responsible to produce an object code suitable for execution using a stack (called run time stack). In other programming languages such as FORTRAN, COBOL, BASIC, which do not have a dynamic memory management mechanism, it is the user's responsibility to define and maintain the stack in order to implement the recursive definition of the procedure. For an illustration, let us consider the calculation of 5! using the recursive definition. To do this, the following steps are needed:Stacks 125 Here, it is required to push the intermediate calculations till the terminal condition is reached. In the above calculation for 5!, Steps 1 to 6 are the push operations. Then subsequent pop operations will evaluate the value of intermediate calculations till the stack is exhausted. To control the recursion, we have to maintain the following stacks: Stack(s) for parameter(s) // To store the parameter(s) with which the recursion is defined Stack(s) for local variable(s) 1/ To hold the local variable(s) that are used within the definition A stack to hold the return address. From the above list of stacks it is evident that the number of stacks required is as many as there are parameters in the procedure, For some procedure which does not use any local variable, no stack is required for that purpose. Similarly, a stack to hold the return address may not be required for some recursive procedure, Details about implementing recursive procedures using a stack are illustrated in the following sections with some well-known problems. Let us assume that all stacks are represented using an array structure. Also, assume that the sizes of stacks are adequate to run a procedure. We will illustrate the implementation of three popular recursive computations: 1, Calculation of factorial value 2. Quick sort 3. Tower of Hanoi problem For each problem, we will deseribe the recursive description, then the translation of the recursive description to a non-reeursive version using stacks. 454 Factorial Calculation Recall that the factorial for an integer N can be defined recursively as follows:126 Classic Data Structures Factorial(N) Steps: 1. If(\= 0) then 2. fact=1 3. Else 4, fact = W * Factorial(W - 1) 5. Endit 6. Return (fact) 7. Stop To implement the above, we require two stacks: one for storing the parameter N and another to hold the return address. No stack is necessary to store local variables, as the procedure does not possess any local variable. Let these two stacks be PARAM (for parameter) and ADDR (for return address). We assume PUSH(X, ¥) operation to push the items X and Y into the stack PARAM and ADDR, respectively. Algorithm FactorialWithStack Input; An integer N, and MAIN, the address of the main routine, say. Output: Factorial value of N (that is N!). Data structure: Array representation of stack. ‘Steps: 1. val = N, top = 0, addr = Step 15 2. PUSH (val, addr) IV Initialize the stack 3. val = val - 1, addr = Step 11 4 Next value and return address 4. If (val = 0) then 5. fact =1 6. Goto Step 12 7. Else 8. PUSH(val, addr) Mf Nal pushed into PARAM and addr pushed into ADDR 9. Go to Step 3 10. Endif 11. fact = val * fact 12. val = POP_PARAM(), addr = POP_ADDR() 13. Go to addr 14, Return (fact) 15, Stop Note that Steps 3-10 control PUSH and Steps 11-13 are POP operations. Here, the stacks are initialized by the calling value and address of the Return statement, that is, Step 14. POP_PARAM() and POP_ADDR() are the two POP operations assumed on two stacks PARAM and ADDR, respectively. The above implementation is strated for N = 5 in Figure 4.6.Stacks 127 ‘Action Execution Step Stack content 1 TOP=0 1 addr = Stop 10 PUSH(S, Stop 10) 2 PA 5 5. Ste v vale, addr = Step7 3 ADDR * 2 val 0 4 5[4 PUSH(4. Step 7) 6 1017 t 3 Vale, acs Step7 3 e]aTa PUSH(3, Stop7) 6 tol 7 t 4. vake2, acdeSt0p7 3 STaTale PUSH(2, Step7) 6 wol7l7l[7 f 6. valet, addr=Step7 3 val 0 4 5 Sy2it PUSH(1, Stop 7) 6 10 7|7\|7 ¥ 6 vale, addt=Step7 : 4[3f2|[1 8 wl7{7[7]7 F a 8 5 a]2|s T 10 tI7\|7 FX 8 & s[4[3]2 T wl7[7]7 tx 9. vale3, addr=Step7 5]4]3 fact=3°2(=6) wl7]7 rr 10. valed, addr=Step7 8 fact=4*6{=24) 7 it valeS, addr=Step7 & fact=5*24(=120) 7 10 + x 12. Goto Step? 10 Figure 4.6 Computation of a factorial (recursively) using a stack.128 Classic Data Structures It may be observed that writing an iterative version for the calculation of a factorial is not only easy but its execution is also efficient. The example for factorial calculation using a stack is only of theoretical interest to illustrate the stack mechanism of a recursive procedure execution. The next two examples highlight the actual advantages of stacks in recursion. 45.5 Quick Sort One useful application of a stack is to sort a number of data using the quick sort algorithm, The quick sort algorithm is based on the divide and conquer technique. The principle behind the divide and conquer technique is to divide a problem into a number of sub-problems. Again each sub-problem is divided into a number of smaller sub-problems and so on till a sub-problem is not decomposable. Solving a problem means solving all the sub-problems. In the case of quick sort, the list to be sorted is partitioned into two sub-lists so that sorting these two sub-lists is sorting the main list; sorting the sub-list again follows the same procedure recursively. Note that partition is to be done in such a way that sorting of the sub-lists is the sorting of the original list. The question is how can this be done. One simple idea is to select any element in the list (let it be the first element and be termed pivor element), Place the pivot element on the list so that all the elements before the pivot element are smaller and all the elements after the pivot element are larger than it, As an illustration, let us consider the following list of numbers to be sorted: 9 65 35 21 48 59 87 52 ; Here, 41 is selected as the pivot element, which is encircled. In order to place 41 in its right position, first compare this element with the element at the extreme right end (shown with an upright arrow below it, we call this the pointer), swap the elements if they are not in order (that is, if the element at the extreme right end is smaller than the pivot element), otherwise move the pointer to left one step and repeat the comparison. In the given list, we see that the pivot element is greater than the element at the extreme right end and hence they are swapped. The list after the swap operation is shown below: 28 719 65 35 a 48 59 87 52 t We see that this swap places the pivot element at the extreme right end. We now compare the pivot element with the element which is next to the element just swapped, see the pointer). In this case, a swap will occur if the comparison tells that they are not in order (that is, the pivor element is smaller than the element on the left), otherwise move the pointer to right one step and repeat the comparison. In the above list, the comparison leads to a swap operation and the list after the swap appears as shown below. 28 65 35 21 48 so 87 s2 79Stacks 129 Repeatedly applying the above steps one can get the following observations, Comparisons with the pivot will shift the pointer to the left side from 52 to 87, 87 to 59, 59 10 48 and 48 to 21, when we find that the pivot element is larger than 21. So a swap will take place: 28 21 65 35 @ 48 39 87 52 19 t Comparing from left, we get the following change: ~ oo G@) s 6 8 © # 2 t Another comparison from right yields 28 21 35 65 48 59 87 52 19 Now, we should stop as we have scanned the entire list (as the pointer has reached the pivot) and we observe that 41 is now placed in its final position, and we get two sub-lists, one containing all the elements to the left of 41 which are smaller than 41 and another list to the right of 41 containing elements which are larger than 41. 28 21 35 This explains how we can partition a given list into two sub-lists. It may be noted that the two sub-lists are not necessarily of equal length. After getting the two sub-lists, we have to apply the same procedure on each sub-list, Then a sub-list is divided into two other sub-lists and the repetition continues until there is a sub-list containing no elements or a single element (this is the terminal stage of the repetition over a sub-list), Basically, the repetition is in a recursive manner, Now, here is the application of a stack, Whenever a sub-list is divided into two sub- lists, one sub-fist has to be pushed onto the stack before sorting the other list. When taking care of this list, pop the neat list to be considered and the procedure will continue till the stack is empty. The quick sort algorithm now can be defined recursively as follows; Let L be the original list and FL, EL be the locations of the front and end elements of L, respectively. Algorithm QuickSort Input: L is the list of elements under sorting with FL and EL being the two pointers at the two extremes. Ouiput: The list L is sorted in ascending order. Remarks: List is in the form of an array and the algorithm is defined recursively.130 Classic Data Structures 9. 10. i. If (EL-FL ¢1) then —_// Termination check: List is empty or contains a single element Exit 11 Terminate the QuickSort for the list Endif loc = Divide(FL, EL) 1 Partition the list into wo sub-lists If (loc = FL) > 1 Af the leftmost list L contains more than one element QuickSort(FL, loc—1) JI Apply quick sort on the leftmost list Endif If (ELoc) > 1 #7 If the rightmost sub-list contains more than one element QuickSort(loc+1, EL) Wf Apply quick sort on the rightmost list EndIf Stop Note that a sub-list can be identified by the location of its two extreme elements. Here, SizeQf{L) is assumed to determine the number of elements in L. Before going to describe quick sort using a stack, let us assume the following: (i) The list of elements is stored in an array A. (ii) In order, to push a list onto the stack we actually push the locations of the first element and the last element; hence two stacks are required. Let these two stacks be represented using two arrays, namely FRONT and END. The quick sort algorithm uses a procedure Divide(). Let us define it next. Algorithm Divide Input: FL and EL are boundaries, that is, the locations of the front and end elements of the list to be divided. Ouiput: LOC is the location of the pivot element which is between the two sub-lists after the divide. Data structure: Acray representation of a stack with TOP as the pointer to the top-most element, Steps: 1. left = FL, right =EL Initialization: left and right are two pointers at the extremes 2. loc = FL MH loc denotes the location of the pivot 3. While (loc # right) and (Afloc] < A[right]} do // Compare from right, pivot is being at left 4. right = right-1 Move to the left 5. EndWhile 6. If (loc = right) then 1 List is scanned fully or list contains a single element 7. Return (loc) 11 Element is placed in its final position 8. Else Mf Elements are not in order and henee swap 9. Swap (Affoc), Alright]} Interchange the pivot and the element on the right of it 10, left = lott If Set the left marker 1. loc = right 1 New position of the pivot clement 12, Endlt (ContdStacks 131 13. While (loc # left) and (Alloc] 2 Alleft}) do // Compare from left pivot is being at right 14, left = left +1 # Move to the right 15, EndWhi 16. If (loc = left) then 1 List is fully scanned as it contains a single element 17. Return(loc) H Element is placed in its final position 18, Else 1 Elements are not in order and swap 19, Swap (Afloc], Afleft) WH Interchange the pivot and the element on the left of it 20, right = loc — 1 Wf Set the right marker 21. toc = left H New position of the pivot element 22. Endif 23. Go to Step 3 JJ Repeat the steps of scanning 24. Stop Now, we shall define the quick sort algorithm. Algorithm QuickSort Input: An array A with N clements Ouiput: Sorted list of elements in A in ascending order Data structure: Array representation of stack with TOP as the pointer to the top-most element. 1. flelel=N 1 Boundaries of the list, 2. top = NULL, HY Stacks are empty initially 3. WWW> 1) JAE the list is not empty or has more than a single clement 4. PUSH(A, el) 1] Push the values into their respective stacks 5. EndIf 6, While (top # NULL) do 11 Till the stack is not empry 7. POP(, el) If Pop a sub-list from stacks %& Divide (fi, el, loc) —_/ Divide the list into two subsists and get the position of pivot 9. IE (f< loc = 1) then /f Test for left sub-list whether it has more than one element 10. PUSH(HL, loc - 1) // The left sub-list is large enough and will be considered later i. Endle 12, If (el > loc +1) then! Test for right sub-List whether it has more than one clement 12. PUSHloc + 1, el) J/ The right sub-list is large enough and will be considered later 4. Endl 1s. EndWhile 16, Stop An illustration of recursive execution of quick sort using stacks is shown in Figure 4,7. In the above algorithm QuickSort( ), we assume PUSH(FL, EL) to push FL and EL into the FRONT and END of stacks, respectively. Similarly, POP(FL, EL) is to POP the items from two stacks FRONT and END and they are stored as FL and EL, respectively. The details of the quick sort method are illustrated through an example as shown in Figure 4.7. Note that an132 Classic Data Structures TOP 4 5 6 7 8 9 10 eronr| 41 21| 48] so 52 | 268 REAR | 10 input ist Stack (2) Initially the entire list (1, 10) Is Pushed Into the stack 1239 4 5 6 7 8 8 1 rpont! | za] 21 | 25 [9] 65] 48 [50] 07 | 52] 70 REAR —u— > (©) The list (1, 10) is POPed and divided Into two sub-Hsts L1(1, 3) and L2(5, 10) Using the Divide procedure and pushed into the stack TOP 4 0 then. #1. = O is the termination condition 2. Move(W - 1, ORG, DES, INT) 3. ORG — DES (Move from ORG to DES) Move (N — 1, INT, ORG, DES) 5. Endit 6. Stop BY For N = 3, how will this recursion solve the problem as shown in Figure 4.9. Move (1, A,B,C) A=»C Move (2, A, C, B) AsB- AaB oa NN Move (1, G,A,B)--- CB Mova (3, A, 8, C) ASC Asc Move (1, 8,0.) B+A Move (2, B, A, C) BC oe Bsc MoveE (1,A,8,C) A+o Figure 4.9 Tower of Hanoi (with N = 3) solution with recursion. Now, let us implement this recursion using stacks, For this purpose, we have to assume the following stacks. STN is to store the number of discs STA is to store the pillar of origin STB is to store the intermediate pillar ste is to store the destination pillar STADD for the return address PUSH, X. ¥, Z, R) and POP (N, X, Y, Z, R) are the two stack operations over these stacks and are expressed as follows: PUSHIN, X, Y. Z, R) POP(N, X, ¥, Z, R) TOP = TOP +1 STN[TOP] = N STA[TOP] = X STC[TOP] = Z STADD[TOP] = & TOP = TOP -1 With these assumptions, the recursive implementation of the tower of the Hanoi problem will be defined as follows:Stacks__ 135 Algorithm HanoiTower Input: N = number of discs, A = origin, B = intermediate and C = destination pillar. Output: Steps of movements. Data structure: Array representation of a stack 1 # Initially all the stacks are empty 2 . des = (Cn =N J} Initialization of the parameters 3. HW Retum to the end step 4, PUSH(n, org, int, des, add) Mf Push the initial value to the stacks 5. If (STNftop] = 0) then I Terminal condition reached 6. Go to STADDItop] 7. Else Mf Translation of move (N = 1, 4, C, BY 8 n= STN[top] = 1 9. org = STALtop] 12. addr = Step 15 11 After completing these moves return to Step 6 13, Go to Step 4 14, Endif 15. POP(n, org, ini, des, r) 16. Print “Move disc from:” org — des Mf Move the nth disc from A to C 17. Do the following: 18, n= STN{top] 19. org = STB{top} 20. int = STAftop] 2. des = STCItop] 22, addr = Step 24 23. Go to Step 4 24. POP(n, org, int, des, r) 25. Gotor # Return address 26. Stop 1 Jt Translation of move (N - 1, B, A, O) To verify the algorithm HanoiTower, the reader can trace down the steps for N = 1, N = 2. N= 3, and N= 4 discs. It may be observed that the minimum number of moves required with N discs is 2% = 1. | Assignment 4.5 | The celebrated Fibonacci sequence (denoted as Fo, Fy, Fx, F5. ==, F,) is a8 below: | 11,263, 5, 8 13. ‘The aboye sequence can be defined using recursion as follows: In =0orm= 1 then F,= 1136 Classic Data Structures Else F, = Fy + Fro Show how this Fibonacci sequence can be generated by writing a recursive program and show how recursive implementation can be done using a stack Assignment 4.6 To compute the greatest common divisor (GCD) of two integers M and N, the recursive definition (called Euclid’s algorithm) as follows: GCD(N. M) if N>M GCD(M, y= M, ifN=0 GCD(N, MOD(M, N)), otherwise Here MOD(M, N) is the module function which retums the remainder when M is divided by N. ‘Write a recursive program for the above implementation using a stack. 4.5.7 Activation Record Management Before discussing the application of stacks to the management of activation records, let us say a few words about its theory. The block structured (also called procedural) programming language allows a user to define a number of variables having the same name or different names in various blocks. The scope of a variable is defined as the regions (that is the blocks) over which the variable is accessible. For example, consider the block structured program depicted in Figure 4.10. Reter2:X Figure 4.10 A block structured program. Here, the variable X is declared in block A as well as in block C. Now references of the variable name at locations say, Refer 1 and Refer 2, are corresponding to which declaration? ‘This can be decided by a scope rule. There are two scope rules: the static scope rule and the dynamic scope rule. The static scope rule defines the scope of a name in terms of the syntacticStacks 137 structure of a program. This rule is called ‘static’ because one can determine a variable’s definition by looking at the program text alone. The static scope rule can be defined as below: © The scope of a variable declared in a particular block consists of that block, exclusive of any block nested within it that declares the same identifier. © Ifa variable is not declared within a block, then it obtains the declaration from the next outer block, if not there then the next outer block, and so on w a declaration is found. Such a rule is known as the ‘most closely nested rule’. Thus, with this rule, reference of a variable at Refer 2 (see Figure 4.10) will be resolved from the declaration in block A, whereas reference at Refer 1 will be resolved from the declaration in block C. In the dynamic scope rule, on the other hand, reference of an identifier is resolved during the execution of the program and the same variable name may be defined at several points within the same program, that is, the variable name may change its definition as the execution proceeds. This is why the dynamic scope rule is also termed ‘fluid binding’. This rule is stated as follows. @ The declaration of a variable is referred from the most recently occurring and still active definition of the name during the execution of the program. For example, consider the program structure shown in Figure 4.11. P is the main program. During its execution, at a certain point, the procedure ‘call A’ occurs. Here, procedure A is in the currently active block P. So, for any reference, X in A will be obtained from the declaration in P. For the other procedure ‘eall2 C’, as it itself has declaration for X (Declare 3) so any reference of X will be resolved from that only. Now, suppose B is on execution. B in turn calls Procedure A (as ‘call4 A’). Here, reference of X will be obtained from Declare 2 as B is the most recently occurring block. Thus, for a given reference of X in A, its declaration is once resolved from Declare! and another from Declare 2 in the same program. Pp Declare 1:X calls C Refer 1:X Declare 2: X calla A Declare 3: Refer2:X TE allt A call2 C calla B Figure 4.11 Dynamic scope rule.138 Classic Data Structures Assignment 4.7 Let us consider the following program structure: MAIN Boolean b = TRUE Procedure P. Bogin ‘Output (b) End Procedure Q Boolean b = FALSE Begin Output (6); call P; End BEGIN IMAI call P; call Q; END What output will you find (a) if the scope rule is static and (b) if the scope rule is dynamic. Implementation of scope rules using stack The next question that arises is how to implement a scope rule. This is completely a burden ‘on the system programmer. Actually, the implementation of scope rules is meant to solve the problem of allocation of memory variables that are declared in different blocks. As we know, there are two different types of storage allocations, namely static storage allocation and dynamic storage allocation. The static storage allocation is easy to implement and efficient from the execution point of view. Here, all variables which are required for a program are allocated during compile time. This is why static storage allocation is known as a compile time phenomenon. In this scheme, each subprogranvsubroutine of a program is compiled separately and the space required for them is reserved till the completion of execution of the program. The space required for a program is, thus, just the sum of the space needed for the program and the subprograms—the space never changes as the program is running. On the other hand, in dynamic storage allocation, the space for memory variables is allocated dynamically, that is, as per the current demand during the execution. When a subprogram is invoked, space for it is allocated and the space is returned when the subprogram completes its execution. Thus, the space required to run a program is not fixed as in static allocation; rather it varies as the program is executed. We are to discuss the implementation of scope rules using the dynamic memory allocation strategy. Easy implementation is possible using a stack called a run time stack. In thisStacks 139 implementation, when a subprogram is invoked, a block of memory required for it is allotted and as soon as the execution is completed it is freed. A single chunk of storage, called an activation record, is used for this purpose. An activation record typically contains the following information: Storage for variables local to the subprograms. Declaration of the procedures and pointers (address of the starting location) to the definitions of procedures in the subprogram. @ The return address (after the end of subprogram, where the control should return). © A pointer to the activation record of the location (the location of the block to. which the subprogram belongs). Thus, for the above-mentioned information, the structure of an activation record can be represented as shown in Figure 4.12. Storage for local variables, Declaration of Procedures and their starting addresses Retum address: Pointer to the location Figure 4.12 Structure of an activation record. A stack for storing the activation records needs to be maintained during the execution of program (note that this stack should be with list structure because of the dynamic nature of the programs). ‘When the program control enters a new subprogram, its activation record is pushed onto the stack and when the subprogram finishes its execution, the control returns to an address which can be obtained from the field ‘Return Address’ of the activation record and this activation record is removed from the stack by updating the stack pointer. For example, for a program as shown in Figure 4.13, whete A is the main program, it invokes B, B in tum invokes C and D. When a subprogram finishes its execution, then the next subprogram to be resumed can be decided by maintaining a stack. Next, we will see, how the scope of a memory variable can be resolved. To do this, let us first consider the pseudo code of a program, as listed in Figure 4.14,‘A program structure Figure 4.13 Execution of program and its run-time stack, a1 oz 03 ot os 06 or 08 9 10 u R B “ 15 16 vv 18 9 20 21 2 23 24 oy 26 a 28 29 30 31 2 33 4 Stack AB ABC AB ABD AB ‘Action ‘Activation record of ‘Ais pushed A invokes B B calls G ‘G completes execution B calls D D completes execution Return to, Rur-tima stack Program MAIN A,B,C sinteger Procedure Q—— Begin AZAD c=CH2 End Procedure R CC: integer Begin c=2; call B=A+B End ——____] Procedure S —— B,C: integer; Procedure Q callR call S End Figure 4.14 A block structured program.Stacks 141 This program requires a total of five activation records. These are displayed in Figure 4.15. MAIN a A,B,C: integer (4), R70) s(22) ot {+ potum aadress ej} + Return adcress MAIN(1) — @-| @-| + Address of the cater ‘avalon record Tar ‘aivaion record for MAIN procedure Qin MAIN R s C :intager B,C: integer ae) 0} Return address 01+ Rotum addross o} > Address of the o} > address of the — caller caller ‘Adlvation record for ‘Adlvalion record for procedure R in MAIN procedure $ in MAIN ais) | —+ Retumm address ©-|—+ dross of tho ‘i ‘halvationrecordtor = SMT procedure Oia S igure 4.18 Activation records of various procedures. However, the field for the pointer to the locator depends on the scope rule. For the dynamic scope rule, there is no need to maintain such a field. Now, during the execution, the activation records of various procedures are either to be pushed or to be popped to and from the stack as per the requirement of the program. We will consider the stack represented with a single linked list, whose node structure is as shown in Figure 4.16,142 Classic Data Structures Pointer to the activation record, 1 e Pointer to the ‘ next node Figure 4.16 Node structure for linked list representation of a stack. Implementation of static scope rule The reference of a variable will be resolved by consulting the current activation record, if it is not resolved here then it will be resolved from the activation records of its caller, and so on. Here the caller means a program/subprogram which calls the subprogram under discussion. The run-time stack view during the execution of the program MAIN (Figure 4.14) is illustrated in Figure 4.17. ?r MAIN A=? Be? *] C=? Qf4) AC10) S(22) Line 94 (a) MAIN begins its execution at line 28, Its activation record is PUSHed into the stack TOP 4 Ly beh a CH? Ant, B-2, 0-3 Ka) F(10) S22) Line 33 Line 34 eo (b) Program control reaches the line 32, Fils invoked and its activation record is PUSHed into the stack Figure 4.17 Continued.Stacks 143 TOP . ‘O(Matn) MAIN C22 A-9,BA2, 025 os - O(a) R10) S22) | Line 13 Line 33 Une 34 (©) Procedure R begins its execution at line 10 and invokes procedure Q al line 12. As Qiis not in A, 80 trom its pointer to the caller it is resolved and the corresponding activation record of Cis pushed. Execution of Q begins al line 4. References of A and G (at line 5 and 6) are resolved from MAIN, the outer block of Q. TOP e}---—- + R t MAIN (4) (10) $(22) Line 33 Line 34 a eo (3) When Q finishes its execution, control gets the retum address from its activation record which is line 13. Activation record of Q is removed. Now, references to B ‘and A at ine 13 are obtained trom their declaration in MAIN, TOP ole tT 8 t MAIN Ba?C=? And, Bab, O=5, Qa) Q(4) R(10) S(22) Line 34 Line 24 aa Si {e) When R completes its execution, control retums to line 93; the procedure S is invoked whose reference is resolved by the current activation record namely, MAIN and record of S is PUSHed into the stack Figure 4.17 Continued,144 Classic Data Structures TOP t e}—____ e|___ 2 J oe Ly OKs) + s +t MAIN [*) Ax3, Be5, C=5 aus) 4) R10) $(22) Line 26 Line a4 Line 94 eH 4 (f) Procedure & begins its execution at line 22 and when control reaches the lino 25 it involves the procedure Q. The reference of Q is resolved from the current activation record, that is, of $ and then the activation record of @ is then PUSHed into the stack. TOP ( m4 phos Ge ED wn m B=3C=2 As4, Be5, CaS ase) aya) ero) s(22) pore rere ren = = (g) Execution of @ is started and references of A and ¢ at lines 19 and 20 ‘are resolved from $ and MAIN, respectively. ! ce? B=30=2 Ax4,B=5, Cx5. a(18) | (4) (10) S(22)} Line 27 line 34 Line 34 {h) When Q finishas its execution, control retums to line 26, its activation is then POPed ‘and procedure R is invoked. This A is resolved from the activation record of MAIN Figure 4.17 Continued.Stacks 145 [els {¢ ls {ole tg + O(MAIN) t R t 8 MAIN c=2 B=9C=2 AeA, — (18) (4) C10) (22) Line 13 Line 27 Line 34 Line 34 + eo (i) R begins its execution at line 10. Reference of C is resolved from the activation record af R. Itagain invokes for ts activation record, FIs searched, which in tum searches the activation record of MAIN; $0 the raference of Q is resolved from MAIN. tls ° $ + (MAIN) + s MAIN cee | Bas0ue [As6, Ba5, Ox? - | 18) la(4) FUt9) Si2a) Line 13, Line 27 Line 34 Line 34 (Ce se La {) Q starts exacution at line 5; references of A and C at lines 5 and 6, respectively, are resolved from MAIN. TOP + s tH MAIN Ax6, B11, C7 2¢4) F(10) S122) Line 34 (k) When Q finishes its execution, control returns to line 13, Q's activation record is retumed; ‘current activation record ig R. References of A and B (at ine 13) are resolved from MAIN. When A finishes its execution at line 14, cantrol gets the ratum address the line 27, Riis removed; control next returns to line 27. Line 27 is the end of S, 80 $ is completed; control returns to line 34. Line 34 is the end of the program MAIN. The execution of the program reaches its end. Figure 4.17(a}{k) Execution of MAIN using the static scope rule.146 Classic Data structures Implementation of dynamic scope rule Implementation of the dynamic scope rule is much easier thn the implementation of the static scope rule. For the dynamic scope rule, the structure of an activation record is the same as for the static scope male except that here it is not required to maintain a pointer field to store the address of the locator. On reference of a variable, its declaration will be searched first from the current activation record; if not found then the next activation record on the stack and so on till the declaration is found or all the records on the stack are searched, As in the static scope rule, here also the execution of a subprogram starts with pushing its activation record onto the stack and, when the execution is finished, the activation record is simply wiped out (that is popped). For the program structure as mentioned in Figure 4.14, its execution using the dynamic scope tule is illustrated in Figure 4.18. (a) R10) $(22) (a) Main begins its execution at line 28. its activation record is PUSHed into the stack. vr + t MAIN c=? A=1,B=2,.0=3 (4) FA(10) S(22) Line 33 Line 36 (b) During the execution of MAIN when control reaches line 22 and call of R occurs, the activation record of R is PUSHed into the stack. Figure 4.18 Continued.Stacks 147 TOP 4 ac ¢ |e “Le q Q 1 R MAIN Cx2 A=1,B=2,C=3 Q(4) Ri10) (22) Line 13 Line 33 Line 34 (6) Controt reaches line 12, the execution of Q is initiated. This Q will be referred from the first activation record present in the stack, that is, trom MAIN, TOP [+le}—— eh}——13 } Q R MAIN wats QM) RCO) (22) Lino 13 Line 33 Line 34 (d) When © begins its execution at line 4 the reference of A is from MAIN and that of C is from A. O(a) (10) S(22) Line 33 Line 34. (@) Qcompletes its execution, control returns to line 13; A.and B are relerred from the activation record of MAIN. B is updated. Top 4 s ¢ t s t MAIN 8=70=2 ASB, B5, 0-3 auiey (4) RCV0) S122) Line 34 Line 34 {f) When R finishes its execution, contro! retums to line 33; call of S occurs ‘and the activation record of S is PUSHed into the stack. Figure 4.18 Continued.148 Classic Data Structures TOP 4 ra — Ga — ars) + s {MAIN B=3C A=3, BeS, Cx3 Gia) FTO) amie) S22) Line 26 Line 94 Line 94 (9) Execution of S begins at line 22 (referred from MAIN) and references of B and C are trom ‘the activation record of S. When control reaches line 25, invocation of Q accurs. Reference of Qiis resolved from the first occurrence; that is, rom the activation of Sin stack. TOP t 19 +} ars) tf s {MAIN B=3G=2 Asd, Bs5, C=3 atts) Oe FeO Line 26 Line 34 Line 34 (h) Q begins its execution at line 18. References of A and C (at lines 19 and 20, respectively) vill be resolved from MAIN and 5, respectively. TOP ‘ ¢ [+--+ [>| ——+12 } R t s {MAIN c=? B=3C=2 ‘zd, Bo5, C=3 ‘Oy ALTO os) s(22) ine 27 Line 34 Line 34 {)) finishes its execution and control reaches line 26, Fils initiated. its activation record is PUSHed into the stack, control jumps to line 10. Top : -e}—_—(e9 1s} 9X : R t 8 _} MAIN C22 B=30-2 ‘Az4, BoS, C=3 ata) oy Line 27 Line 34 Line 34 () During the execution of R, C is reselved trom the activation record of A. Next, when Q is. invoked, reference of Q is resolved from the first occurrence, that is, from S. The Qis at line 18. Figure 4.18 Continued,Stacks 149 TOP * Le |e Le |e + ais) tT a yt s MAIN C=3 B=3, C=2 (A=5, B=5, C=3 ‘Q(4) (10) Q(4) R(10) aca) Oy” ey” Line 13 Line 27, Line 34 Line 34 {k) During the execution of Q, reference of A is referred from MAIN arid that of C is from TOP 4 tle Le lx ain C=3 A=5,B=5,C=3 (4) RIO) ‘$(22) Lino 27 Line 3¢ () After Q finishes its execution, control retums to line 13. For the reference of B and A at line 13, Bis resolved for S and Ais from MAIN. Later, when the execution of Fis completed, control returns to line 27, which indicates that the execution of S is finished. then control returns to line 34 which is the end of the program MAIN. Figure 4.18 (a)-(I)_ Execution of MAIN using the dynamic scope rule. Assignment 4.8 Consider the following recursive function: Function Fibonacci (n: integer) Begin If (n = 0) or (m= 1) then fibo = 1 else fibo = Fibonacci (n — 1) + Fibonacei (n — 2) End Suppose the size of an activation record required for function Fibonacci (n) is as follows: For local variables and passed parameter = 4 bytes For return address = 2 bytes For the node structure (on stack) it requires 4 bytes. If the above function is run on a computer with a stack of 640 bytes, estimate the maximum value of # for which the memory will not underflow, Give reasons for your answer.150 Classie Data Structures Assignment 4.9 Consider the following program: Program TEST Procedure A — Procedure B- Assuming suitable activation records for various procedures, draw the display structure just after the procedures: marked as X, and ) marked as ¥ are called. Make sure to indicate which of the two procedures named A you are referring to. 46 PROBLEMS TO PONDER 4.1 Suppose we need to maintain two stacks of the same type of items in a program. If the two stacks are stored in separate arrays, then one stack might overflow while there are considerable unused spaces in the other. To avoid this situation it is better to maintain the two stacks in the same array as shown in Figure 4.19. Stack A Stack B Bottom A Top Top 8 Bottom B Figure 4.19 Double stack structure.Stacks 151 In this structure, one stack, say A, grows from one end of the array and the other stack, say B, starting from the other end, grows in the opposite direction, that is, towards the direction of A. Write algorithms as well as the C++ program of the above-mentioned double stack structure for the following operations: PUSHA, PUSHB, POPA and POPB. 4.2 Repeat the same concept as in Problem 4.1 but for three stacks in an array. 4.3 Draw a schematic diagram showing how to plan six stacks in a single array. 4.4 The efficient method used in evaluating a polynomial of the form P,Q) = age! + ax) + ay? 4 4 ae + a, is by nesting using Homer's rule, as shown below: a6) = (0% (Cdk + ay) + age EO + ayy) “OR + Oy Show how this can be carried out using a stack. 4.5 Consider the following arithmetic expression in postfix notation: 752+*415-/~ (a) Find the value of the expression. (b) Find the equivalent prefix form of the above expression. (¢) Find the value of the expression from its prefix notation. 4,6 Devise a method to convert an infix expression into its prefix form that includes (, ), +. ah 4.7. Write a method to convert a postfix expression to infix for the same set of symbols as in Problem 4.6. 4.8 An arithmetic expression is given in postfix form; it is required to convert it into its equivalent prefix form. Write a method for this (Hint: You can use the method as obtained in Problems 4.6 and 4.7 and combine them suitably.) 4.9 Add the exponentiation operator to your repertoire for Problems 4.6, 4,7 and 4.8. 4.10 In our discussion, we have only been concerned with binary operators. In an arithmetic ‘expression, there are also some operators like plus (+) and minus (-) which can be used both for binary as well as unary. Repeat Problems 4.6, 4.7 and 4.8 assuming unary plus (+) and unary minus (~) operators in the expression, (Hint: It is an easy matter to distinguish different occurrences of unaries, an operator denotes a binary operator if it does not occur either at the beginning of an expression or immediately after a left parenthesis.) Consider the following arithmetic expressions: (@) AS BS CD (b) A* BIC 41 =152 Classic Data Structures () -A+B-C/A forA=2,B=3,C=-4, (d) (A+B) *C+D/(B+A*C)+D (©) ABAC+D*E-A+C Convert the above expression into (i) Postfix notation (ii) Prefix notation 4.12 List all the prime factors of the given integers in descending order. (Hint: Use stack.) 4.13 Devise a method that will produce all permutations of the first N integers using stack. 4.14 There is a variation in the original Euclid’s algorithm for computing the greatest common divisor of two integers M and N. According to the modified Euclid's algorithm, GCD(M-N,N) ifM2N GCD(M,N)=5 ifN=0 GCD(M,N-M) if N>M Using only a stack, write a procedure to compute the GCD as per the modified Eu: method. REFERENCES Bruno, J. L. and T. Lassagne, The generation of optimal code for a stack machine, Journal of the ACM, July 1975. Donald E. Knuth, The Art of Computer Programming, Vol. 2, Addison-Wesley, Reading, Massachusetts, 1984. Forsythe, A.L, T.A. Keenan, et al., Computer Science: A First Course, John Wiley & Sons, New York, 1986. Harrison, M.C., Data Structures and Programming, Glenview, Berkeley, California, 1985, Sethi, Rajeev and J. Ullman, The generation of optimal code for arithmetic expressions, Journal of the ACM, Vol. 10, October 1970.UcUues 5.1 INTRODUCTION A queue is a simple but very powerful data structure to solve numerous computer applications. Like stacks, queues are also useful to solve various system programs. Let us discuss some simple applications of queues in our everyday life as well as in computer science before undertaking the study this data structure. Queuing in front of 2 counter Suppose there are a number of customers in front of a counter to get service (say, to collect tickets or to withdraw/deposit money in a teller of a bank), Figure 5.1(a). The customers arc forming a queue and they will be served in the order they arrived, that is, a customer who comes first will be served first. Counter Figure 5.4(a) Queue of customers. 153154 Classic Data Structures Traffic controt at a tuming point Suppose there is a turing point in a highway where the traffic has to turn, See Figure 5.1(b). All the traffie will have wait in a line till it gets the signal for moving. On getting the ‘Go’ signal the vehicles will tum on a first come, first turn basis. 4 2882 @) aes —- Figure 5.1(b) Traffic passing at a turning point. Process synchronization in multi-user environment In a multi-user ehvironment, more than one process is handled by the monitor (opetating system). See Figure 5.1(c). The three different states that a process may have are the following: READY, RUNNING, and AWAITED. A process is in the READY state when it is submitted to the system for execution. A process is in the RUNNING state if it is currently under execution, Similarly, a process will be transferred to the AWAITED state when it requires resource(s) which is/are busy. In order to synchronize the execution of processes, the monitor has to maintain two queues, namely QU and Q2, for READY and AWAITED states respectively where a process which entered a queue first will be exited first. (eon) ‘oz = eS Ready Awaited Figure §.1(c) Queues of processes. Resource sharing in a computer centre In a computer centre, where resources are limited compared to the demand, users must sign a waiting register. See Figure 5.1(d). The user who has been waiting for a terminal for the longestQueues 155, period of time gets hold of the resource first, then the second candidate, and so on. Here the waiting list maintains a queue and the first signed will be the first allowed. my man Fe +— waiting register Figure 5.1(d) A waiting queue of users in a computer centre, 5.2 DEFINITION Like a stack, a queue is an ordered collection of homogeneous data elements: in contrast with the stack, here, insertion and deletion operations take place at two extreme ends. A queue is also a Jinear data structure like an array, a stack and a linked list where the ordering of elements is in a linear fashion. The only difference between a stack and a queue is that in the case of stack insertion and deletion (PUSH and POP) operations are at one end (TOP) only, but in a queue insertion (called ENQUEUE) and deletion (called DEQUEUE) operations take place at two ends called the REAR and FRONT of the queue, respectively. Figure 5.2 represents a model of a queue structure, rT T Enqueua— | @ |e) = | = | —» dequeue Figure 5.2 Model of a queue. An element in a queue is termed ITEM; the number of elements that a queue can accommodate is termed LENGTH. From the examples mentioned in Section 5.1 and the definition as stated above, it is evident that a data in a queue is processed in the same order as it had entered, that is, on a first-in, first- out basis. This is why a queue is also termed first-in first-out (FIFO).156 Classic Duta Structures 5.3. REPRESENTATION OF QUEUES There are two ways to represent a queue in memory: 1. Using an array 2. Using a linked list The first kind of representation uses a one-dimensional array and it is a better choice where a queue of fixed size is required. The other representation uses a double linked list and provides a queue whose size can vary during processing. The following two subsections describe the represcntation of queues in memory. 5.3.1 Representation of a Queue using an Array ‘A one-dimensional array, say Qf ... NJ, can be used to represent a queue. Figure 5.3 shows an instance of such a queue. With this representation, two pointers, namely FRONT and REAR, are used to indicate the two ends of the queue. For the insertion of the next element, the pointer REAR will be the consultant and for deletion the pointer FRONT will be the consultant. Front Rear Figure 5.3 Array representation of a queue. Three states of a queue with this representation are given below: Queue is empty FRONT = 0 REAR =0 (and/or) Queue is full REAR = N FRONT = 1 (when full by compact) Queue contains elements 2 1 FRONT ¢ REAR Number of elements = REAR - FRONT + 1 Now let us define the operation ENQUEUE to insert an element into a queue. Algorithm Enqueue Input: An element ITEM that has to be inserted. Quiput: The ITEM is at the REAR of the queue. Data structure: Q is the array representation of a queue structure; two pointers FRONT and REAR of the queue Q are known.Queues 157 1 Queue is full 2 3 4 3. If (REAR = 0) and (FRONT = 0) then H Queue is empty 6 FRONT 7 Endif 8 REAR =REAR +1 1/ Insert the item into the queue at REAR 9. O[REAR] = ITEM 10, Endif 11. Stop ‘The deletion operation DEQUEUE ean be defined as given below: Algorithm Dequeue Input: A queue with elements. FRONT and REAR are the two pointers of the queue Q. Output: The deleted clement is stored in ITEM. Data structures: Q is the array representation of a queue structure. Steps: 1. If (FRONT = 0) then 2. Print “Queue is empty” 3. Exit 4. Else 5. ITEM = Q[FRONT] 11 Get the element 6. If (FRONT = REAR) # When the queue contains a single element 7. 8. 9. REAR = 0 It The queue becomes empty . Else 10. FRONT = FRONT + 1 LL. Endif 12, Endl 13, Stop Let us trace the above two algorithms with a queue of size = 10, Suppose the current state of the queue is FRONT = 8, REAR = 9. Ten operations are requested as under; 1, DEQUEUE 2. ENQUEUE 3. ENQUEUE 4, DEQUEUE 5. DEQUEUE 6, DEQUEUE 7. ENQUEUE 8. ENQUEUE 9, DEQUEUE 10. DEQUEUE Figure 5.4 presents the status of the queue when these operations are carried out.158 Classic Data Structures Queue at its current state 1 2 3 4 5 6 7 6 8 1 tT FOR 1. Request : DEQUEUE > tf 2. Request : ENQUEUE. FR > FF 3, Request : ENQUEUE FoR > Message Queue is full t t 4, Roquest : DEQUEUE > Tf 5. Request : DEQUEUE FR ths FR 6. Request : DEQUEVE empty 9. Request : DEQUEUE tt < FR 10, Request : DEQUEUE wo LITT) Figure 5.4 Operations on a queue.Queves 159 ‘There is one potential problem with this representation. From Figure 5.4, we can see that with this representation, a queue may not be full, still a request for insertion operation may be denied. For example, on request (3) (in Figure 5.4) 8 rooms are available but insertion is not possible as the insertion pointer reaches the end of the queue. This is simply a wastage of the storage. This type of representation can be recommended for an application where the queue is emptied at certain intervals. Assignment 5.1 The algorithm Engueue may fail even though there is memory space available. One way to- avoid this problem is to rewrite the algorithm Engueue and Dequeue. Two solutions are suggested as given below: Suggestion 1; Rewriting the algorithm Enqueue. Whenever the REAR pointer gets to the end of the queue (Figure 5.5), test whether the pointer FRONT is at location 1 or not; if not, shift all the elements so that they are wrapped from the beginning and thus make room for a new item. Suggestion 2: Rewriting the algorithm Dequeue. t + a FF & Figure 6.5 Shifting elements when R reaches to the end. a+ After the end of each deletion, all the elements at the trail are shifted once towards the fror here the idea is to fix the FRONT pointer always at 1, The queue which follows such operations is termed a dynamic queue. Rewrite operations ENQUEUE and DEQUEUE for a dynamic queve. 5.3.2 Representation of a Queue using a Linked List One more limitation of a queue, other than the inadequate service of insertion represented with an array, is the rigidness of its length. In several applications, the length of the queue cannot be predicated before and it varies abruptly. To overcame this problem, another preferable representation of a queue is with a linked list. Here, we select a double linked list which allows us to move both ways. Figure 5.6 shows the double, linked list representation of a queue. The pointers FRONT and REAR point the first node and the last node in the list. Roar | [fe [para <4 [oatan Figure 5.6 A double linked list representation of a queue.160 __ Classic Data Structures Twa states of the queue, either empty or containing some elements, can be judged by the following tests: Queue is empty FRONT = REAR = HEADER HEADER-—RLINK = NULL Queue contains at least one element HEADERSRLINK # NULL The insertion and deletion operations are straightforward and the same as in the algorithm InsertEnd_DL (for Enquewe) and algorithm DeleteFront_DL (for Dequeue); these two algorithms are already defined in Section 3.4 of Chapter 3. Assignment 5.2 Explore the possibilities of representing a queue using an ordinary single linked list or circularly single linked list. 5.4 VARIOUS QUEUE STRUCTURES So far we have discussed two different queue structures, that is, either using an array or using a linked list (and a variation of a queue structure using an array as an assignment). Other than these, there are some more known queue structures. This section discusses them. 5.4.1 Circular Queue ‘As pointed at the end of Section 5.3.1, for a queue represented using an array when the REAR pointer reaches the end, insertion be denied even if room is available at the front. One way to avoid this is to use a circular array. Physically, a circular array is the same as an ordinary array, say A[I ... MJ, but logically it implies that A[1] comes after A[N] or after A[N], A[J] appears. Figure 5.7 shows logical and physical views of a circular array. (a) Circular queue (gical) 4) Circular aray (physical) Figure 5.7 Logical and physical views of a circular queue.Queues 16% The principle underlying the representation of a circular array is as stated below: Both pointers will move in a clockwise direction. This is controlled by the Mop operation; for example, if the current pointer is at i then shift to the next location will be i Mop LENGTH + 1, | <¢< LENGTH (where LENGTH is the queue length). Thus, if i= LENGTH (that is at the end), then the next position for the pointer is 1. With this principle the two states of the queue regarding, i.e. empty or full, will be decided as follows; Circular queue is empty FRONT = 0 REAR = Circular queue is full FRONT = (REAR mop LENGTH) + 1 The following two algorithms describe the insertion and deletion operations on a circular queue, Algorithm Enqueue_CQ Input: An element ITEM to be inserted into the circular queue. Output: Circular queue with the ITEM at FRONT, if the queue is not full. Data structures: CQ be the array to represent the circular queue. Two pointers FRONT and REAR are known. Steps: 1. If (FRONT = 0) then 11 When the queue is empty 2. FRONT =1 3. REAR=1 4. CQIFRONT) = ITEM 5. Else 11 Queue is not empty 6. next = (REAR mop LENGTH) + 1 7. If (next # FRONT) then 17 If the queue is not full 8. REAR = next 9. CQIREAR] = ITEM 10, Else uM. Print “Queue is full” 12 Endif 13. Endlf 14, Stop Algorithm Dequeue_CQ Input: A queue CQ with elements. Two pointers FRONT and REAR are known. Output: The deleted element is ITEM if the queue is not empty. Data structures; CQ is the array representation of circular queue.162 Classic Data Structures Steps: 1, If (FRONT = 0) then 2. Print “Queue is empty” 3. Exit 5. ITEM = CQIFRONT] 6. If (FRONT = REAR) then #7 If the queue contains a single element 7. FRONT = 0 8% REAR = 0 9. Else 10. FRONT = (FRONT Mop LENGTH) + 1 le Endif 12. Endif 13. Stop In order to trace these two algorithms, let us consider a circular queue of LENGTH = 4. The following operations are requested. Different states of the queue while processing these requests are illustrated in Figure 5.8. 1. ENCQUEUE (A) . ENCQUEUE (C) . DECQUEUE . DECQUEUE . DECQUEVE » DECQUEUE Feawy 2. ENCQUEUE (B) 4, ENCQUEUE (D) 6. ENCQUEUE (E) 8. ENCQUEUE (F) 10, DECQUEUE 12, DECQUEUE Assume that initially the queue is empty, that is, FRONT = REAR = 0. Initially the stack is empty (1) Request : ENCQUEUE (A) tt LTT 42) Request : ENQUEUE (B) (8) Request ; ENQUEUE (C) (4) Request : ENQUEUE () A B c T t < i A < A ele D B}ec|o f T. t } F a > F a Figure 5.8 Continued.Queues 163 (6) Request : ENQUEUE (E) (?) Flequest : DEQUEUE E c|o E clo AF R F (@) Request ; ENQUEUE (F) (9) Request ; DEQUEVE e|rijc]o E|F D f t t 7 ROOF R F* (10) Request : DECOUEUE (11) Request : DECOUEUE e|F F T TT FOR < FR < (12) Request : DECQUEUE jt — Queue is empty tt FR Figure 5.8 Tracing insertion and deletion operations on a circular queue, Assignment 5.3 The following two algorithms are proposed for a circular queue. Algorithm 1 Enqueuel_CQ Steps: 1. If (REAR + 1) mop’ LENGTH = FRONT) then 2, Print “Queue is full” 3, Exit 4, Else 5. CQ[REAR} = ITEM 6 1 8. . REAR = (REAR + 1) sop LENGTH . Endl . Stop Algorithm 2 Dequeue2_CQ Steps: 1. If (REAR = FRONT) then 2, Print “Queue is empty” 3. Exit 4. Else (Contd)164 Classic Data Structures 5. ITEM = CQ(FRONT] 6, FRONT = (FRONT + 1) Mop LENGTH 7. Endit 8. Stop {a) Decide the queue that fits with these algorithms. (b) Do the two algorithms utilize properly the whole of available memory? If your answer is ‘NO’, then devise the necessary modification so that this deficiency is overcome. 54.2 Deque Another variation of the queue is known as deque (may be pronounced ‘deck’). Unlike a queue, in deque, both insertion and deletion operations can be made at either end of the structure. Actually, the term deque has originated from double ended queue. Such a structure is shown. in Figure 5.9. FRONT REAR Deletion ——> ——+ Deletion Insertion =—— Ineertion Figure 5.9 A deque structure, It is clear from the deque structure that it is a general representation of both stack and queue. In other words, a deque can be used as a stack as well as a queue. ‘There are various ways of representing a deque on the computer. One simpler way to represent it is by using a double linked list. Another popular representation is using a circular array (as used in a circular queue). ‘The following four operations are possible on a deque which consists of a list of items: 1, Push_DQUITEM): To insert ITEM at the FRONT end of a deque. 2. Pop_DQ(): To remove the FRONT item from a deque. 3. Inject(ITEM): To insert ITEM at the REAR end of a deque. 4, Eject(): To remove the REAR ITEM from a deque. ‘These operations are described for a deque based on a circular array of length LENGTH, Let the array be DQ[I ... LENGTH). Algorithm Push_DQ Input: ITEM to be inserted at the FRONT. Output: Deque with a newly inserted element ITEM if it is not full already. Data structures: DQ being the circular array representation of a deque.Queues 165 Steps: 1, If (FRONT = 1) then 1 1€ FRONT is at extreme left 2 ahead = LENGTH 3. Else Md If FRONT is at extreme right or the deque is empty 4, If (FRONT = LENGTH) or (FRONT = 0) then 5. ahead = 1 6. Else 7. ahead = FRONT - 1 11 FRONT is at an intermediate position 8, EndIf 9. If (ahead = REAR) then 10. Print “Deque is full” UL Exit 12. Else 13. FRONT = ahead 11 Push the ITEM 14. DQIFRONT] = ITEM 15. Endif 16. Endlf 17. Stop Algorithm Pop_DQ /* This algorithm is the same as the algorithm Dequewe_CQ *f Algorithm Inject /* This algorithm is the same as the algorithm Enqueue_CQ */ Algorithm Eject_DQ Input: A deque with elements in it. Output: The item is deleted from the REAR end. Data structures: DQ being the circular array representation of deque. Steps: |. If (FRONT = 0) then Print “Deque is empty” . Exit }. Else If (FRONT = REAR) then # The deque contains single clement ITEM = DQIREAR] #1 Deque becomes empty 4 REAR is at extreme left ITEM. = DQ[REAR] REAR = LENGTH Else If (REAR = LENGTH) then 1 REAR is at extreme right ITEM = DQ[REAR]16. Else df REAR is at an intermediate position 17. ITEM = DQIREAR] 18. REAR = REAR = | 19. Endle 20. Endlf 21. EndIf 22. EndIf 23. Stop There are, however, two known variations of deque: 1. Input-restricted deque 2. Output-restricted deque. These two types of variations are actually intermediate between a queue and a deque. Specifically, an input-restricted deque is a deque which allows insertions at one end (say REAR end) only, but allows deletions at both ends. Similarly, an output-restricted deque isa deque where deletions take place at one end only (say FRONT end), but allows insertions at both ends. Figure 5.10 represents two such variations of deque. FRONT REAR — Insertion Deletion —+ -+— Deletion Input-restricted deque 7 ~ Insertion——e +— Insertion Deletion «— (0) Output-restricted deque Figure 5.10 Types of deque. Assignment 5.4 . Using the linked list representation of a deque, obtain the four operations PUSHQQ, POPQ(), INJECT(), and EVECT(). . Using the circular array representation of a deque, obtain the following: v (a) Insertion operation into an input-restricted deque. (©) Deletion operations from an input-restricted deque. . Repeat problem 2 for an output-restricted deque. »Quewes 167 5.4.3 Priority Queue A priority queue is another variation of queue structure. Here, each element has been assigned a value, called the priority of the element, and an element can be inserted or deleted not only at the ends but at any position on the queue. Figure 5.11 shows a priority queue. FRONT REAR at { Py} pa] -[>l-[- Pn Figure 5.11 View of a priority queue. With this structure, an element X of priority p; may be deleted before an element which is at FRONT, Similarly, insertion of an element is based on its priority, that is, instead of adding it after the REAR it may be inserted at an intermediate position dictated by its priority value. Note that the name priority queue is a misnomer in the sense that the data structure is not a. queue as per the definition; a priority queue does not strictly follow the first-in first-out (FIFO) principle which is the basic principle of a queue. Nevertheless, the name is now firmly associated with this particular data type. However, there are various models of priority queue known in different applications, Let us consider a particular model of priority queue. 1. An element of higher priority is processed before any element of lower priority. 2. Two elements with the same priority are processed according to the order in which they were added to the queue. Here, process means two basic operations namely insertion or deletion. There are various ways of implementing the structure of a priority queue. These are: (Using a simple/circular array (ii) Multi-queue implementation Gii) Using a double linked list (iv) Using heap tree. We will now see what cach of these implementations is. (Heap tree implementation of priority queue will be discussed in Chapter 7 of this text.) Priority queue using an array With this representation, an array can be maintained to hold the item and its priority value. The element will be inserted at the REAR end as usual. The deletion operation will then be performed in either of the two following ways: (a) Starting from the FRONT pointer, traverse the array for an element of the highest priority. Delete this element from the queue. If this is not the front-most element, shift all its trailing elements after the deleted element one stroke each to fill up the vacant position (see Figure 5.12),168 Classic Data Structures FRONT REAR <+—Insartion ato =—— Deletion i Shittone Figure 5.12 Deletion operation in an array representation of a priority queue. This implementation, however, is very inefficient as it involves searching the queue for the highest priority element and shifting the trailing elements after the deletion. A better implementation is as follows: (b) Add the elements at the REAR end as earlier. Using a stable sorting algorithm”, sort the elements of the queue so that the highest priority element is at the FRONT end. ‘When a deletion is required, delete it from the FRONT end only (see Figure 5.13). vv “ x Y Deletion « Insertion ~N ‘Sort after ineerton_—— Figure 5.13 Another array implementation of a priority queue. The second implementation is comparatively better than the first ones here the only burden is to sort the elements. The algorithms of the above two implementations are left as assignments to the reader. Multi-queue implementation This implementation assumes W different priority values. For each priority p; there are two pointers F; and R; corresponding to the FRONT and REAR pointers respectively. The elements between F; and R; are all of equal priority value p,. Figure 5.14 represents a view of such a structure. Fy Fy Ay F R t fog + je= 20 Figure 5.14 Multi-queue representation of a priority queue. With this representation, an clement with priority value p; will consult F; for deletion and R; for insertion. But this implementation is associated with a number of difficulties:Queues 169 (® It may Jead to a huge shifting in order to make room for an item to be inserted. ii) A large number of pointers are involved when the range of priority values is large. In addition to the above, there are two other techniques to represent a multi-queue, which are shown in Figures 5.15(a) and 5.15(b). It is clear from Figure 5.15(a) that for each priority value a simple queue is to be maintained. An element will be added into a particular queue depending on its priority value. The priority queue as shown in Figure 5.15(b) is in some way better than the multi-queue with multiple queues. Here one can get rid of maintaining several pointers for FRONT and REAR in several queues. A multi-queue with multiple queues has one advantage that one can have different queues of arbitrary length. In some applications, it is seen that the number of ‘occurrences of elements with some priority value is much larger than the other value, thus demanding 2 queue of larger size. i i Priowity 2, FE R 12 LENGTH 1 i | 1 . . 3 Prony Pe EEE preva? priority value 4 (or Prionty 2, N Nx LENGTH (a) Multiple quoue with simpto queues (b) Mutiple queue with a matrix Figure 5.15 Mult-queue implementation with multiple simple queues and matrix. Both the above representations are not economic from the memory utilization point of view: much of the memory space remains vacant. Algorithms for insertion and deletion operations for multi-queue implementation are left as exercises for the student. Linked list representation of a priority queue This representation assumes the node structure as shown in Figure 5.16, LLINK and RLINK are two usual link fields, DATA is to store the actual content and PRIORITY is to store the priority value of the item, We will consider FRONT and REAR as two pointers pointing the first and last nodes in the queue, respectively. Here all the nodes are in sorted order according to the priority values of the items in the nodes. The following is an instance of a priority queue: *A sorting algorithm is stable if the relative positions of two identical items remain the same in the unsorted and sorted list (for details, see Chapter 10).170 Data Structures i LLINK | DATA |PRIORITY| RLINK ef e100 OFEEECEELEGR REAR Figure 5.16 Linked list representation of a priority queue. FRONT With this structure, to delete an item having priority p, the list will be searched starting from the node under pointer REAR and the first occurring node with PRIORITY = p will be deleted. Similarly, to insert a node containing an item with priority p, the search will begin from the node under the pointer FRONT and the node will be inserted before a node found first with priority value p, or if not found then before the node with the next priority value, The following, two algorithms Insert_PQ and Delete_PQ are used to implement the insertion and deletion operations on a priority queue. Algorithm Insert_PQ Input: The ITEM and its priority P value of a node that is to be inserted. Ouput: A new node inserted. Data structures: Linked list structure of priority queue; HEADER as the pointer to the header. Steps: 1. ple = HEADER WF Start from the first node 2. new = GetNode(NODE) H Avail anew node 3. newDATA = ITEM HI Get initialized the node with ITEM 4, new-»PRIORITY = P 5. While (pt—+RLINK + NULL) and (ptrPRIORITY < P) do // Search for the position 6. 1. 8. ptr = ptr-aRLINK . EndWhile . If (ptr—eRLINK = NULL) then —_// If the list is empty or the item is with the largest Priority value 9. puraRLINK = new 10. new-»LLINK = pir J The node is inserted as the last node (Conteh)Queues 171. 14, If (ptr>PRIORITY 2 P) then #1 First occurrence is found 15. pirl = ptr—>.LLINK MP Ansert the new node 16. ptrlRLINK = new 1 Before the node with priority > P 17. new—RLINK = ptr 18. ptr>LLINK = new 19. new—LLINK = pirl 20. Endl 21. Endlf 22, FRONT = HEADERSRLINK 1 Set the FRONT pointer 23. Stop 1. 2. 3. 4, 5. 6. 7. 8. Similarly, the algorithm for deletion can be described as follows: Algorithm Delete_PQ Input: The priority P of the clement which has to be deleted Qutput: The element that is being deleted. Data structures: Linked list structure of priority queue; HEADER as the pointer to the header. Steps: If (REAR = NULL) then Print “Queue is empty” Exit Else ptr = REAR ‘While (ptr>PRIORITY > P) and (ptr + HEADER) do pir = ptr-oLLINK End While If (ptr = HEADER) or (ptr>PRIORITY < P) Print “No item with priority”, P Exit Else If (ptu—priority = P) then If (ptr = REAR) IIE the last node to be deleted REAR = pirl ptrl--RLINK = NULL Else 4 Other than last node pirlsRLINK = ptr 1 Deleted ptr2—2LLINK = ptrl Endit Endit (Contd)172 Classic Data Structures 24. Endif 25. item = pt-»DATA 26. ReturnNode(item) 27. Endit 28. Stop Assignment 5.5 (a) Describe the insertion and deletion algorithms for the following priority queue structures: () Queue represented with a single large array. (ii) Multi-queue representation with a single large array. (iii) Multi-queue representation with multiple simple queue. (iv) Multi-queue with matrix representation. (b) The algorithms Insert_PQ and Delete_PQ assume that the highest priority element will be deleted first. Modify the two algorithms so that the lowest priority element will be deleted first. 55 APPLICATIONS OF QUEUES Numerous applications of queue structures are known in computer science. One major application of queues is in simulation, Another important application of queues is observed in the implementation of various aspects of an operating system. A multiprogramming environment uses several queues to control various programs. And, of course, queues. are very much useful to implement various algorithms, For example, various scheduling algorithms are known to use varieties of queue structures. This section highlights a few applications and then illustrates how powerful queues are to solve different problems. 5.5.1 Simulation Simulation is modelling of a real-life problem, or in other words, it is the model of a real-life situation in the form of a computer program. The main objective of the simulation program is to study the real-life situation under the control of various parameters which affect the real problem, and is a research interest of system analysts or operation research scientists. Based on the results of simulation, the actual problem can be solved in an optimized way. Another advantage of simulation is to experiment the danger area. For example, areas such as military operations are safer to simulate than to field test, simulation being free from any risk as well as inexpensive. Simulation is a classical area where queues can be applied. Before discussing simulated modelling, let us study a few terms related to it. Any process or situation that is to be simulated is called a system. A system is a collection of interconnected objects which accepts zero or more inputs and produces at least one output,Queues 173 (see Figure 5.17(a)). For example, a computer program is a system where instructions are the interconnected objects and inputs or initialization values are the inputs and the results obtained during the execution constitute the output. Similarly, a ticket reservation counter is also a system (you can easily guess about its input, output and functionality). Note that a system can be composed of one or more smafler system(s). ‘A system can be divided into different types as shown in Figure 5.17(b). A system is discrete if the input/output parameters are of discrete values. For example, a customer arriving at a ticket reservation counter is a discrete parameter, whereas water flowing through a pipe to a reservoir is an example of a continuous system since the parameter is of the continuous type. Interconnected INPUT——+| objects to provide | + output a functionality (a) Outing of a system SYSTEM t Discrete Continuous Deterministic Stochastic ‘system ‘system system system (b) Classification of systems Figure 5.17 System and simulation. A system is deterministic if from a given set of inputs and initial conditions of the system, the final outcome can be predicted. For example, a program to calculate the factorial of an integer is a deterministic system. On the other hand, a stochastic system. is based on randomness: its behaviour cannot be predicted before hand. As another example, the number of customers waiting in front of a ticket reservation counter at any instant cannot be forecasted. There may be some systems wl are intermixes of both deterministic and stochastic. After getting an idea about the various types of systems, let us define various kinds of simulation models. There are two kinds of simulation models: event-driven simulation and time- driven simulation: these are decided according to how the state of a system changes, In the case of time-driven simulation, the systems changes in its states with the change time, and in event- driven simulation, the system changes its state whenever a new event reaches the system or exits from the system. Now let us consider a system, its model for simulation study and then the application of queues in it, Consider a system as a ticket selling centre, There are two kinds of tickets available, namely T1 and T2, which customers are to purchase. Two counters Cl and C2 are available (Figure 5.18). Also assume that the time required for issuing a ticket of TL and T2 are f, and fy, respectively. Two queues QL and Q2 are possible for the counters C1 and C2, respectively. With this description of the system, two models are proposed:174 Classic Data Structures Model 1 © Any counter can issue both types of tickets. A customer on arrival joins the queue which has a lesser number of customers; if both queues are equally crowded, then to Ql, the queue of counter Cl. Model 2 © Two counters are earmarked, say C1 for selling T1 only and C2 for selling T2 only. © A customer on arrival goes to either queue QI or queue Q2, depending on whether the ticket T] or T2 is to be purchased. Figure 5.18 A ticket selling counter. To simplify the simulation model, the underlying assumptions are made: 1. Queue lengths are infinite. 2. One customer in a queue is allowed to purchase one ticket only. 3. Let 4, and A, be the mean arrival rates of customers for tickets TI and T2, respectively. The values for 4, and A, will be provided by the system analyst. 4. Let us consider the discrete probability distribution (also called Poisson distribution) for the arrival of customers to the centre, Poisson distribution gives a probability function P@) =1-e# where P(t) = the probability that the next customer arrives at time 1, and 4 = the mean arrival rate, Thus, if we assume N to be the total population of customers in a day, then N= APW= Nd =e") 6.1) is the number of customers who arrived at the centre for ticket TI at time f, and N= NPC) ) is the number of customers who arrived the centre for ticket T2 at 5. A clock is maintained with an initial value (to dictate the opening and closing of counters) when the counter is made available to the customer, etc.176 Glassic Data Structures With these basic assumptions and definitions, the proposed simulation model may be termed the discrete deterministic time-driven simulation model, Before describing the algorithms, let us assume the following keywords that will be assumed in our algorithm: 1. 6. An abstract data type CUSTOMER to simulate a customer with the following components: CUSTOMER @ TicketReq—the customer requesting for the type of ticket © 4,—1ime of arrival of the customer © fy—time of departure of the customer @ r—amount of time served when a customer is currently under service: . SETCUSTOMERI( ) and SETCUSTOMER2( ) are the two functions which set the data for TicketReq as T1 and T2, ¢as 4 and t for the object of CUSTOMER. 7; in each case will be set by the current value of the CLOCK. . LENGTH(Q) is a function to return the length of a queue Q. |. ENQUEUE(CUSTOMER) and DEQUEUE( ) are the two functions to add a customer into the queue and to remove a customer from the queue, respectively. SEECUSTOMER(Q) function is to attend a customer which is at front of a queue Q by the counterman. This function is actually to read the Ticket Req and 1 for a customer C. EMPTY(Q) retums “TRUE’ if the queue Q is empty else ‘FALSE’. Now, we are in a position to write the algorithms for Modell and Model2, which are stated as follows: Algorithm SimulationModel_1 Inputs: A, = Mean arrival rate of customers for ticket T; A, = Mean arrival rate of customers for ticket T; N = Total population assumed OPEN_HOURS = Total time duration for which the centre remains open in a day f; = Time required to process for a ticket T; 4, = Time required to process for a ticket Tx Outputs: Agi = Average queue length of the queue Q) Ag, = Average queue length of the queue Qy Ty, = Average waiting time of the customers for ticket Ty Ty, = Average waiting time of the customers for ticket T; Tg, = Total service time that the counter C, served Tc, = Total service time that the counter C; served Data structures: Two queues Ql and Q2 represented with double linked lists, An abstract data type CUSTOMER.Queues 177 Steps: /* INITIALIZATION */ 1. Q) and Q are initialized as empty 2. CLOCK to maintain the current time is set at zero 3. L.=1,=0 J These are required for average lengths of Q, and Qz 4. WT, = WT; 11 For storing the waiting times of customers for T, and T; 5. NT, = NT, =0 H! Number of customers served with tickets T, and T, 6 NC\T, = NC\T; = 0 J) Number of customers served in counter C, for ticket T), Tp 7. NCsT; = NC3T, =0 J) Number of customers served in counter C3 for ticket T}, Ts START SIMULATION FOR A DAY*/ 9. While (CLOCK < OPEN_HOURS) do #*Generate population of customers*/ 10. Ny) =N* (1 -e%* CLOCK) MW. N; =N * (1 —e®* CLOCK) /*Add customers into the queues*/ 12. i = Length(Q,) 1 Current lengths of the queue Q, 13. & = Length Q:) 1 Current lengths of the queue Qy 14, AF (N, > Ny) then 15. While (N; > 0) do 16. ¢ = SetCustomerl() 4 Customer for T, 17. Tf (!, $4) then 18. Enqueuel(c) M1 Add to the smaller queue 19. hehel 20. Eke 2. Enqueue2 (C) 22, h=h+tl 23 Endif 24, M=N-1 25. ¢ = SetCustomer? () #{ Customer for T2 26. If (i, £ [,) then H Add to the smaller queue 27. Enqueuel (c) 28. hehel 29. Else 30. Enqueuel (c) 31. h=h+1 32. EndIf 33. N= 2-1 34. EndWhile (Contd)Classic Data Structures ‘While (N, > 0) do // Add the remaining customer for ticket of type T ¢ = SetCustomerl () If ($f) then H Add to the smaller queue Enqueuel (c) hahel Else Enqueue2 (c) h=h+1 EndIf M=N-1 End While Else H Customers for T, are more than those for T,, that is, Nj >My ‘While (N, > 0) do /f One customer from T, type and other from T; type ¢ = SetCustomerl () If (iy 0) do ¢ = SetCustomer2 () If d, <4) then Enqueuel (c) hehel Enqueue2 (c) babel (Contd)179 76. 7. 79. 81. 82. 83. 85. 86. 87. 88. 89, 92, 93. 9s. 96. 97. 98. 100. 101. 102. 103. 104. 105, 106, 107. 108. 109. 110, Endlt Ny=Nz-1 EndWhile Lal, +h, =k +h /Servicing the customers in queues*/ /*AT COUNTER Cy*/ If (EMPTY(Q,) = TRUE) then Go to Step 108 Else c = SeeCustomer(Q1) If (e - TicketReq = T,) then If (c— 1 > 0) then c-tscerH Else ¢-2= CLOCK WT, = WT, + (€ e-l NC,T, = NC,T, +1 NT, = NT +1 Dequeuel() Endif Else If (c= ¢ > 0) then c-tec-t-1 Else c= & = CLOCK WT, = WT, + (e- R-e-1) NT; = NT +1 NC\T, = NC\T2 + 1 Dequeuel() Endif Endlf /*AT COUNTER C2*/ If Empty(Q,) then Go to Step 135 Else c¢ = SeeCustomer (Q:) If (¢- TicketReq =7,) If (c- > 0) then 14 Sum up the lengths of queues for average H# Look for the customer in front of Q, A The customer for ticket type T, JC, processed one customer of T, # The customer for ticket type T; calculation MAE not finished It Service for one unit df Service finished M# Time of completion iNWaiting time for customer of T; 14 One customer of T, is served H Leave the customer for Q; (Conid.)180 Classic Data structures ll. 112. 113. ¢- = CLOCK 14. WT, = WT, + (c- @-c- a) 115. NCoT, = NCaT + 1 116. NT) = NT +1 17. Dequeue2() 118. Endl 119. Else 120. If (c- 1 > 0) then 123. 124. 125. 126. 127. 128. 129. 130. 131. 132. 133. 134. 135. CLOCK = CLOCK + | 136. _EndWhile If Blapse one time unit ‘COMPUTE RESULTS*/ 137. Ig; = L/OPEN_HOURS If Average queue length of the queue Qy 138. Iq) = LalOPEN_HOURS W Average queue length of the queue Qo 139. Tp, = WTIOPEN_HOURS #/ Average waiting time of the customers for ticket T; 140. Ty, = WT2/OPEN_HOURS // Average waiting time of the customers for ticket T; 141. Pq, = CT, - f+ NO\T2= H Total service time that the counter Cy served M42. Pq = NCsT,- t+ NCyT2—f2 Total service time that the counter C2 served 143. Stop As stated earlier, the only difference between Model 1 and Model 2 is that in the case of Model 1, a customer either for ticket T, or T> is allowed in any one of the queues, whereas, in the case of Model 2, queues are earmarked, that is, queue Q, is only for the customers of ticket T, and queue Q is only for the customers of ticket T;. From the implementation point of view, Model 2 is simpler than Model 1. The algorithm for Model 2 is as follows: Algorithm SimulationModel_2 Input: Inputs are same as in the algorithm SimudarionModel_I Ouput: Outputs are same as in the algorithm SimulationModel_ Data structures: Same as in the algorithm SimulationModel_1Queues 181 Steps: 2. 20. 2h. 32. 33. /*INTTIALIZATION*? Initialization steps 1.a...1.g remain same as in the algorithm SimudationModel_t F*START SIMULATION FOR A DAY*/ While (CLOCK < OPEN_HOURS) do /*Generate population of customer at the instant CLOCK*/ N= NO -e HO) Ny = N(1= @ PORK) /*Add customers into the queues*/ 4, = Length(Q1) fy = Length (Q2) While (W, > 0) do 11 Add all customers for ticket T, to queue Q, ¢ = SetCustomer!() Enqueuel(c) heel N=M-1 End While While (N > 0) do Mf Add all customers for ticket T; to queue Qz c = SetCustomer2() Enqueue2 (c) SERVICING THE CUSTOMERS*/ #Service at counter C*/ If (Empty(Q,) = TRUE) then Go to Step 34 Else c= SeeCustomer(Q,) If (¢- 1 > 0) then e-tse-t-1 Else ¢- t= CLOCK WT, = WT, + (¢- -e€- 4) NT) =AT, +1 Endif Endif P*Service at counter C3*/ 1/ Length of the queues expands after enqueues (Conic)182 Clascie Data Structures M. If (Empty(Q,) = TRUE) then 35, Go to Step 46 36. Ebe 37. ¢ = SeeCustomer (Q)) 38. If(¢-1>0) then 39. e-tee-r-1 40. Else 41 ¢- = CLOCK 42, WT; = WIp4 (C= R= - tl) 4B. NT; = NT; +1 44, EndIf 45. Endif. 46. CLOCK = CLOCK + 1 47. EndWhile /* COMPUTE RESULTS */ 48. Iq) = L{OPEN_HOURS M1 Average queue length of the queue Q, 49. Igy = L{OPEN_HOURS UV Average queue length of the queue Q, 50. WTL/OPEN_HOURS M Average waiting time of the customers for ticket Ty 51. = WT2 / OPEN_HOURS 1 Average waiting time of the customers for ticket Tz 52. Ter = NCT = 4) + NC\Ta~ ta # Total service time that the counter C, served 53. Ta =NCT— 4 + NCT2- J Total service time that the counter C; served 54, Stop Assignment 5.6 (Simulation of a traffic control system) Let us consider the ease of an automation of a traffic control system. Suppose at a junction point three roads meet (Figure 5.19). For the traffic on the three roads, three signals X. Y and Z are available. Only one signal can be turned on at a time to allow the traffic to pass; for example, if the signal X is on, the traffic from the queue XQ will be allowed to pass either towards Y or towards Z road whereas the traffic in queues YQ and ZQ will wait for their signals. Three models are suggested in order to study the performance of the traffic control: Model 1: Each signal when tumed on allows to pass equal number of vehicles, say, Nin succession. Model 2: Different signals allow different numbers of vehicles during their turn, say, signals X. Y and Z allow N;, N and Ny number of vehicles, respectively. Model 3: Only that signal will be turned on where the maximum number of vehicles are waiting till the number of vehicles waiting are at par with the other remaining queues. If all the queues contain the same number of vehicles, signal X will be turned on. (Conta)Ye za Zz Figure 5.19 A junction of three roads. Assume that if a queue is empty the corresponding signal is not necessary to be turned on. Write simulation programs to obtain a comparative study using the following arrival rate of vehicles. 1. According o Poisson’s distribution 2. According to continuous distribution. 5.5.2 CPU Scheduling in a Multiprogramming Environment In a multiprogramming environment, a single CPU has to serve more than one program simultaneously. This section gives a brief idea about how queues are important to manage various programs in such an environment. Let us consider a multiprogramming environment where the possible jobs for the CPU are categorized into three groups: 1. Interrupts to be serviced. A variety of devices and terminals are connected to the CPU and they may interrupt the CPU at any moment to get a particular service from it. 2. Interactive users to be serviced. These are mainly user’s programs under execution at various terminals. 3, Batch jobs to be serviced, These are long-term jobs mainly from non-interactive users, where all the inputs are fed when jobs are submitted: simulation programs, and jobs to print documents are of this kind. Here the problem is to schedule all sorts of jobs so that the required level of performance of the environment will be attained, One way to implement complex scheduling is to classify the workload according to its characteristics and to maintain separate process queues. So far as the environment is concemed, we can maintain three queues, as depicted in Figure 5.20. This approach is often called multi-level queues scheduling. Processes will be assigned to their respective queues. The CPU will then service the processes as per the priority of the queues. In the case of a simple strategy, absolute priority, the process from the highest priority queue (for example, system processes) are serviced until the queue becomes empty. Then the CPU184 Classi Data Structures switches to the queue of interactive processes which has medium priority, and so on. A lower- priority process may, of course, be pre-empted by a higher-priority arrival in one of the upper- level queues. system Freres | High-priorty queue |__ Evont-driven scheduling Interactive 7 processes: Round Robin f Medium-prlonty queue ing ‘e—* cru FCFS. Within queue: scheduling Figure 5.20 Process scheduling with multilevel queues. Multi-level queues strategy is a genetal discipline but has some drawbacks. The main drawback is that when processes arriving in higher-priority queues are very high, the processes in a lower-priority queue may starve for a long time. One way out to solve this problem is to time slice between the queues. Each queue gets a certain portion of the CPU time. Another possibility is known as multi-level feedback queue strategy. Normally in multi-level queue strategy, as we have seen, processes are permanently assigned to a queue upon entry to the system and processes do not move between queues, The multi-level feedback queue strategy, ‘on the contrary, allows a process to move between queues. The idea is to separale out the processes with different CPU burst characteristics. If a process uses too much of CPU time (that is, long run process), it will be moved to a lower-priority queue. Similarly, a process which is waiting for too long a time in a lower-priority queue, may be moved to a higher-priority queue. For example, consider a multi-level feedback queue strategy with three queues Q), Q) and Qs (Figure 5.21), ay] FoFS Toa, Figure 5.21 A multi-level feedback queue.Queves 185 A process entering the system is put in queue Q,. A process in Q, is given a time quantum fof 10 ms, say. If it does not finish within this time, it is moved to the tail of queue Q,. If Qy is empty, the process at the front of queue Q, is given a time quantum f of 20 ms, say. If it docs not complete within this time quantum, it is pre-empted and put into queue Q,. Processes in queue Q; are serviced only when queues Q, and Q; are empty. Thus, with this strategy, the CPU first executes all processes in queue Q). Only when Qy is empty it will execute all processes in queue Q2. Similarly, processes in queue Q; will only be executed if only queues Q; and Q> are empty. A process which arrives in queue Q, will pre- empt a process in queue Q; or Qs. It can be observed that this strategy gives the highest priority to any process with a CPU burst of 10 ms or less. Processes which need more than 10 ms, but less than or equal to 20 ms are also served quickly, that is, they get the next highest priority over the shorter processes. Longer processes automatically sink to queue Q3; from Qs, processes will be served on a first- come first-serve (FCFS) basis and in the case of a process waiting for too long a time (as decided by the scheduler) it may be put into the tail of queue Q,. 5.5.3 Round Robin Algorithm The round robin (RR) algorithm is a well-known scheduling algorithm and is designed especially for time sharing systems. Here, we will see how a circular queue can be used to implement such an algorithm, Before going to implement the RR algorithm, we should first describe the algorithm with illustration. Suppose, there are m processes P), Ps, ... Py required to be served by the CPU, Different processes require different execution times, Suppose, the sequence of processes” arrivals according to their subscripts, that is, P} comes before P) and, in general, P; comes after P,, for 1 < i Sn. ‘The RR algorithm first decides a small unit of time, called a time quantum or time slice, © A time quantum is generally from 10 to 100 milliseconds. The CPU starts service with Py. P, gets the CPU for time ¢, afterwards the CPU switches to P3, and so on. When the CPU reaches the end of time quantum of P,, it returns to P; and the same process will be repeated. Now, during time sharing, if a process finishes its execution before the finishing of its time quantum, the process then simply releases the CPU and the next process in waiting will get the CPU immediately. As an illustration, consider Table 5.2 for the set of processes: Table 5.2 Table for process and burst time Process P, 7 P; 18 Py 5 The total CPU time required is 30 unit. Let us assume a time quantum of 4 unit. The RR scheduling for this will be as shown in Figure 5.22.186 Classic Data Structures: o 4 68 i 15 19 2 2 2 30 Figure 5.22 RR scheduling. The advantage of this kind of scheduling is reduction in the average turn around time (not necessarily always true). The turn around time of a process is the time of its completion minus the time of its arrival. Thus, using the FCFS strategy, T+(7 +18) +(7 +18 +5) _ 62 ‘Average tum around time * = 20.66 unit Whereas, using the RR algorithm, ‘Average tum around time = Bare = S. 21.66 unit Sce the result by repeating the calculations but using the sequence of processes as Ps, P and P3. In time sharing systems any process may arrive at any instant of time. Generally, all the processes currently under exccution are maintained in a queue. When a process finishes its execution it is deleted from the queue and whenever a new process arrives it is inserted at the tail of the queue and waits for its tum. To illustrate this, let us consider Table 5.3. Table 5.3. Table for process events Process Arrival time Burst time P, 0 9 P; 1 3 Py 9 5 Pa 1 8 The total CPU time required is 25 units. Let the time quantum be r= 5 unit. Figure 5.23 illustrates the snapshot at various instants with RR scheduling. Now let us discuss the implementation of the RR. scheduling algorithm. A circular queue is the best choice for it. It may be noted that it is not strictly a circular queue, because here a process upon completion is deleted from the queue and it is not necessarily from the front of the queue rather it can be from any position of the queue. Except this, RR scheduling follows all the properties of a queue, that is, the process which comes first gets its turn first. The implementation of the RR algorithm using a circular queue is straightforward. Here, we use a variable sized circular queue: the size of the queue at any instant is decided by the number of processes in execution at that instant. Another mechanism is necessary: whenever 4 process is deleted, to fill the space of the deleted process, it is required to squeeze all the processes preceding to it, starting from the front pointer (Figure 5.24). (A detailed procedure for implementation is left as an exercise to the reader.)Queues 187 Pyfinished —P, finished its execution, and deleted P; intial @eleted from trom the in the queue sueue queve Py finished 1 54 wz 7 | P, enters P, ontors P, finished, the queve the queue removed from the queue and ‘the queue is empty Figure 5.23 In and out in a queue during RR scheduling I i are] be proces a deleted IP, le—n Figure 5.24 Deletion of a process from a circular queue. 5.6 PROBLEMS TO PONDER 5.1 A queue is maintained in an array, and F and R are the front location and rear location of the queue, respectively. (a) Obtain a formula for N, the number of elements in the queue in terms of F and R. (b) Write an algorithm to delete the ith element in the queue. (c) Write an algorithm to insert an item X just after the ith clement. $.2 Repeat Problem 5.1 for a circular queue implemented in an array. 53 Write an algorithm REVERSEQ that will reverse all the elements in a queue. [Hint: Use a stack,] 5.4 Write an algorithm REVERSECQ that will reverse all the elements in a circular queue which is maintained (i) in an array, (ii) in a singly linked list. 5.5 It is required to split a queue into two queues so that all the elements in odd positions are in one queue and those in even positions are in another queue, Write an algorithm SPLITQ() to accomplish this, Assume that the queue is maintained in an array. 5.6 Repeat Problem 5.5 to obtain the algorithm SPLITCQ() for a circular queue.188 Classic Data Structures 5.7 A deque is a generalization of both stack and queue. Show how a deque can be implemented using two stacks, (That means two stacks are with you, using stack ‘operations only you have to implement the operations of deque.) 5.8 A generalization of both queue and deque is deck which is stated as “Addition of elemems can be made at the both ends but the deletion can be made either at the end or at the beginning.” ‘Write the operation for deck using the operations of queue and deque only. 5.9 A priority queue can be implemented using a matrix PQ and two arrays FRONT and REAR pointing the front and rear of the queue (Figure 5.25). ‘Obtain the various operations of a priority queue with such an implementation. F FRONT Figure 5.25 A priority queue implementation using a matrix and two arrays. 5.10 Repeat Problem 5.9 with N number of double linked lists instead of matrix PQ and two arrays of pointers FRONT and REAR pointing the front and rear of the lists. Virginia, 1986. REFERENCES Ellis Horowitz and Sartaj Sahni, Fundamentals of Data Structures, Computer Science Press, Rockville, Maryland, New York, 1985. Jean Paul Tremblay and Paul G. Sorenson, Introduction to Data Structures with Applications, McGraw-Hill, New York, 1987. Robert L. Kruse, Bruce P, Leung and L, Clovis Tondo, Data Structures and Program Design in C, Prentice-Hall of India, New Delhi. Thomas L. Naps, Introduction to Data Structures with C, West Publishing Company, West REARTables In this chapter, we will discuss another important data structure called table. This kind of data structure plays a significant role in information retrieval. As an example, suppose that a set of 1n distinct records with keys Kj, Kz, .... K, are stored in a file. We want to find a record with a given key value, K, One simplest way is to perform a sequential search, that is, start from the location of the first record, compare the key K with the key of this record, if found then stop, else proceed to the next record and continue the same procedure. The searching time required is directly proportional to the number of records in the file. If the number of records increases our searching time also increases. However, this searching time can significantly be reduced, even can be made independent of the number of records, if we use a table called access table. ‘This table may store the location of all records in the file, Figure 6.1 shows the use of an access table to retrieve any record in the file storage. Here, we assume a function f, and if this function is applied on X it returns #, an index, so that # = f(K). Then the ith entry in the access table gives us the location of the record with key value K. This method of accessing any record is called table lookup which is, in fact, independent of the number of the records in the file. 189190 Cassie Data Structures Figure 6.1 Information retrieval through table lookup. In this chapter, we will explore the various possible kinds of tables and their usefulness. The different tables of interest are listed below: 1. Rectangular tables 2. Jagged tables 3. Inverted tables 4, Hash tables. The following sections discuss all these tables. 6.1 RECTANGULAR TABLES Rectangular tables are also known as matrices, Matrices have already been discussed in Chapter 2, Section 2.4. Since these tables are needed in various applications, almost all programming languages provide convenient and efficient means to store and access them, so that the programmer does not have to worry about the implementation details. Rectangular tables, therefore, do not require any further discussion. 6.2 JAGGED TABLES Jagged tables are nothing but a special kind of sparse matrices such as triangular matrices, band matrices, eic., which have already been discussed in Section 2.4.2, Chapter 2. In the jagged table, we put a restriction that if elements are present in a row (or in a column) then they are contiguous. Thus in Figures 6.2¢a)-(e), all are jagged tables except the table in Figure 6.2(1). We have seen (in Section 2.4.2) how sparse matrices can be stored in a one-dimensional amay and if the sparse matrices are symmetric in form then how their indexing formula can be decided so that any element in the matrices can be accessed. Recall the indexing formula for any element ay ina matrix is Address (ay) = ixG@-) | Be; 61)Tables 191 5 ® o o Figure 6.2 Jagged tables and sparse matrices. This formula involves multiplication and division which are in fact inefficient from the computational point of view. Here, we will discuss another alternative but improved technique, where we can avoid multiplication and division by setting up an access table whose entries correspond to the row indices of the jagged table, such that the ith entry in the access table is ixG-1) 2 The access table is calculated only once at the time of initiation and can be stored in memory, it then can be referred each time the access of an element in the jagged table occurs. It may be noted that even during the initial calculation, for the entries in the access table, it does Not require any multiplication or division but only addition such as 0,1, (1+ 2) + 2)+3,.. In Figure 6.3, the representation of a jagged table is illustrated. For example, if we want to access as, (the element in the Sth row and the 4th column) then at the Sth location of the access table, we see that the entry is 10; hence the desired element is at 14 (= 10 + 4) location of the array which physically contains the element, It is assumed that the first element of the table is located at location 1 of the array. The above-mentioned accessing technique has another advantage over the indexing formula. We can find an indexing formula even if a jagged table is asymmetric with respect to the arrangement of elements in it. For example, in the jagged table shown in Figure 6.2(d), it is difficult to find an index formula, In this case, however, using an access table, we can easily maintain its storage in an array and can obtain faster access of elements from it.192 Classic Data Structures al [as follows: 1, The function H should be very easy and quick to compute. 2. The funetion H should as far as possible give two different indices for two different key values. As an example, let us consider a hash table of size 10 whose indices are 0, 1, 2, ... 8, 9. Suppose a set of key values are: 10, 19, 35, 43, 62, 59, 31, 49, 77, 33. Let us assume the hash function H is as stated below: © Add the two digits in the key. © Take the digit at the unit place of the result as the index; ignore the digit at the tenth place, if any. Using this hash funetion, the mappings from key values to indices and to hash table are shown in Figure 6.7. In this example, for the given set of key values, the hash function does K [Tt o[t9 to [4 1 [40 19 [0 2| a5 8 a[ a0 43 | 7 4) 50,31, 77 CE) 3 so | 4 ar 4 $8 aa} 73 7[ 43 7_|_4 8| 35,62 33 [6 ® HikoT Hash table Figure 6.7 Example of hashing.196 Classic Data Structures not distribute them uniformly over the hash table; some entries are there which are empty, and in some entries more than one key value needs to be stored. Allotment of more than one key value in one location in the hash table is called collision. We have found three collisions for 62, 31 and 77 in the above-mentioned example. It can be noted that | K | =| J'|, that is, the number of key values is the same as the size of the hash table, but this is not the case always. In general, | K | >| J |. The following are some hash functions which are very common and popularly applied in various applications. Division method One of the fast hashing functions, and perhaps the most widely accepted, is the di which is defined as follows: Choose a number A larger than the number V of keys in K. The hash function H is then defined by ion method, H(k) = K(MOD h) if indices start from 0 or AH(&) = K(MOD A) + 1 if indices start from 1 where k € K, a key value, The operator mop defines the modulo arithmetic operation, which is equal to the remainder of dividing k by A. For example, if k = 31 and h = 13 then H(31) = 31(MOD 13) =5 HI) = 31(MOD 13) +1=6 ‘The number A is usually chosen to be a prime number or a number without small divisors, since this usually minimizes the number of collisions. Generally, h is a prime number and equal to the size of the hash table. Midsquare method Another hash function which has been widely used in many applications is the midsquare method. The method is defined as follows: The hash function # is defined by H(k) = x, where x is obtained by sclecting an appropriate number of bits or digits from the middle of the square of the key value &. This selection usually depends on the size of the hash table. It needs to be emphasized that the same criteria should be used for selecting the bits or digits for all of the keys. ‘As an example, suppose the key values are of the integer type, and we require 3-digit addresses, Our selection criteria are to select 3 digits at even positions starting from the right- most digit in the square. Let us see the address calculations, for 3 distinct keys and with the hash funetion, as defined above: k : 1234 2345, 3456 Po: 1522756 5499025 11943936 Hk): 525 492 933Tables 197 Here, we observe that the second, the fourth, and the sixth digits, counting from the right, are chosen for the hash addresses. The midsquare method has been criticized because of time-consuming computation (multiplication operation), but it usually gives good results so far as the uniform distribution of the keys over the hash table is concerned. Folding method Another fair method for a hash function is the folding method. The method can be defined as follows: Partition the key & into a number of parts ki, ky, Ky where each part, except possibly the las, has the same numberof bits or digits as the required address width. Then the parts are added together, ignoring the last carry, if any. Alternatively, H(R) = hy + ky t= + hy where the last carry, if any, is ignored. If the keys are in binary form, the exclusive-OR operation may be substituted for addition. There are many variations known in this method. One is called the fold shifting method, where the even number parts, k2, k4, ... are each reversed before the addition. Another variation is called the fold boundary method. Here, two boundary parts, namely, k, and &,, each are reversed and then added to all other parts. As an example, let us take the size of each part to be 2; the following calculations are performed on the given key values (integers) as shown below: kz 1522756 5499025 11943936 Chopping: O1 52 27 56 05 49 90 25 11 94 39 36 Pure folding: 01 + 52 + 27+ S6= 136 05+ 49 +90 + 25= 169 11+94+ 39+ 36 = 180 Fold shifting: 10 + 524+ 724 56= 190 50+49+09425= 133 11494493 4 36 =234 Fold boundary: 10 + 52 + 27+ 65 = 154 50 +49+90452= 241 11+94+ 39+ 63 = 207 Folding is a hashing function which is also useful in converting multi-word keys into a single word so that another hashing function can be used on that. In fact, the term “hashing” comes from this technique of ‘chopping’ a key into pieces. Digit analysis method The basic idea of this hashing function is to form hash addresses by extracting and/or shifting the extracted digits or bits of the original key. As an example, given a key value, say 6732541, it can be transformed to the hash address 427 by extracting the digits in even positions and then reversing this combination. For a given set of keys, the position in the ‘keys and the same rearrangement pattern must be used consistently. The decision for extraction and then rearrangement is based on some analysis. To do this, an analysis is performed to determine which key positions should be used in forming hash addresses. For each criterion, hash addresses are calculated and then a graph is plotted, then that criterion is selected which produces the most uniform distribution, that is with the smallest peaks and valleys.198 Classic Data Structures This method is particularly useful in the case of static files where the key values of all the records are known in advance. We have assumed the key values as integers in our previous discussions, but it need not be so always. In fact, any key value can be represented by a string of characters and then ASCII values of its constituent characters can be taken to convert it into a numeric value. Thus, assuming that a key value k = Kyksky ... ky Where each k; is the constituent character in &. The hash funetion using the division method is stated as below in algorithm HashDivision. Algorithm HashDivision Input: K, the key value in the form of a string of characters whose hash address is to be calculated. Outpui: INDEX, a positive integer as the hash address. Data structure: Hash table in the form of an array. H is the size of the hash table which is used for modulo arithmetic operation. Steps: Lisl #7 ‘is the pointer to the string K 2. keyVal = 0 11 To store the keyvalue of K 3. While (K[i] # NULL) do 4. keyVal = keyVal + Kié] Wf Add the ASCH value of K 5 izi4l 1 Move to the next character 6, EndWhile 7. INDEX = keyVal Mop H + 1 11 Find the remainder modulo 8. Return (INDEX) 9. Stop ‘The algorithms for other hash functions can be designed likewise; these have been left as exercises for the student, Assignment 6.2 1. Write the algorithms HashMidSquare, HashFoldingPure, HashFoldingShifi, HashFoldingBoundary, for the midSquare method and different variations of the folding method, respectively, for a given key value K in the form of a string of characters. 2. Suppose a file contains 100 records. What should be the size of the hash table and hence hi? Create an array of key values of the records as hash table. Generate 100 random numbers and assume them as key values. Apply different hash functions to calculate hash addresses and load the key values into the hash table, 3. Generate 100 random numbers each of 6 digits, say. Assuming these as static key values, obtain a digital analysis on it based on three different criteria. Plot a graph for the hash values obtained with these criteria and then select the best criteria.Tables 199 | 4: A hash function will be termed good if it satisfies both the criteria, namely, quick to | compute and uniformity of distribution, The first criteria can be controlled by using a | suitable technique of computation. For example, suppose, for a key k = k,kk3, the hash function is H(k) = k, + ky * 27 + ky * 27. Using Homer's rule for evaluating a polynomial, this can be calculated as: H(K) = (ky* 27 + ka) * 27 + ky Extend the Horner's rule to compute the hash address for the following: Ye aI ist (0) HUikk, +k, Hint: Additions can be replaced by the bit-wise exclusive-OR operation for increased speed. R= > kaa a Hint: Additions can be replaced by the bit-wise exclusive-OR for increased speed and multiplication by 32 is not really a multiplication, it can be done by shifting the bit by 5. (©) A hybrid hash function H'*(K) of H(A) can be obtained by applying the division method over H(k) as stated below: (b) Hlkifok HP(k) = H(k) Mov h h being the size of the hash table. Obtain the algorithm for such a hybrid hash function. 6.4.2 Collision Resolution Techniques Whatever the hash function used in hashing, the complete removal of collisions is almost impossible. This can be emphasized with an example called birth day surprise. Suppose there is a class of 24 students and they are having the same year of birth. We want to know the probability that two students have the same date of birth. The probability can be calculated as follows: Open the calendar of the year of their birth. Assume that there are 365 days. Start with any student, and put a tick on his birthday date on the calendar. Now. the probability that the second student has a different birthday is 364/365. Tick this date off. The probability that a third student has a different birthday is now 363/365. Continuing this way, we see that if the first (# = 1) students have different birthdays, then the probability that the nth student has a different birthday is 365—(n-1) | 365—n+1 365 365 Since the birthdays of different people are independent, we obtain the probability that students all have a different birthday is200 Classic Data Structures 364 363 | 362 x 5a a xl we xe xe xe 365 365 365 365 This probability can be calculated as less than 0.5 whenever n 2 24. In other words, suppose there is a hash table of size 365 and we want to store the records of all the 24 students based on birthdays as their key values. It is therefore a fifty-fifty chance that two of the students have the same birthday and hence a collision. So, collision in hashing cannot be ignored, whatever be the size of the hash table. The next question arises therefore is what to do if their is a collision? There are several techniques to resolve the collisions. Two important methods are listed below: (a) Closed hashing (also called linear probing) (b) Open hashing (also called chaining). 6.4.3 Closed Hashing The simplest method to resolve a collision is closed hashing. Suppose there is « hash table of size h and the key value of interest is mapped to an address location i, with a hash function. The closed hashing then can be stated as follows: Start with the hash address where the collision has occurred, let it be i, Then follow the following sequence of locations in the hash table and do the sequential search. 1 jing cases occurs: BEAN EH Dy ey By Ly 2, one ‘The search will continue until any one of the foll © The key value is found. © An unoccupied (or empty) location is encountered. The searches reaches the location where the search had started. The first case corresponds to the successful search and the last two cases correspond to unsuccessful search. Here the hash table is considered circular, so that when the last location is reached, the search proceeds to the first location of the table. This is why the technique is termed closed hashing. Since the technique searches in a straight line, it is also alternatively termed linear probing; probe means key comparison. Let us illustrate the method with an example. Assume that there isa hash table of size 10 and the hash function uses the division method with remainder modulo 7, namely, F1(k) = k mop (7 + 1), Let us consider the build up of the hash table (initially, the table is empty) with the following set of key values: 15 11 25 1698 128 The loading of the hash table will take place successively by performing a search for a key and inserting it into the table in an empty room if the key is not in the table and leaving if it is overflow, that is, no free room to accommodate any further key value. This is illustrated in Figure 6.8.Tables 201 1 i 1 1 2| 2a{_ 16 |+ 2[_ 15 2[ 75, 3] 3 3| 3| 4] 4 4] 4] 5| 5 ~ |e {os 6 6 6 6|_ 25 7| 7 7 7| a 8 a a 9] 8 3] 3] 19 10 10) 10] Initially the insertion of 15 insertion of 11 insertion of 25 hash table is empty 1 1 1 1 al 15 2|_ 18 als + fis 3{_16_|+ 3|_16_|+ af 16 3|_ 16 4] a{-9 al 9 alo 5H 5 5,1 ~ st 6|_25 6|_25 6|_25 6| 25 |* 7| 7 8 78 | a 8 8 al iz a 9 3 9 19 10 19] 19] insertion of 16 insertion of 9 insertion of 8 insertion of 12 Figure 6.8 Building up a hash table. Next, let us define the operation for searching a key-value and inserting a key-value. The algorithm HashLinearProbe for searching a key value K in a hash table of size HSIZE is given below: Algorithm HashLinearProbe Input: K is the key value of search. INSERT is a flag for the insertion operation. Ouiput: Return the location if it is found in the hash table else if INSERT is TRUE put K in the table if table has not averflown otherwise return NULL. Data structures: A hash table H of size HSIZE in the form of an array. Steps: 1, flag = FALSE I? Flag for continuation of looping 2. index = HashFunetion(K) J} Calculate the hash address using a hash function 3. If (K = Hfindex}) then 17 IE there is a hit 4. Return(index) 5. Exit H End of the execution 6 Else 1 index + 1 HI Set to the next location (Contd202 Classic Data Structures & While (i # index) and (not flag) do 9. If (li = NULL) or (H[i] < 0) then A Af the cell is free 10. If (INSERT) then 11 Truc option for insertion H HU) =K MF But the key value into the hash table 12 flag = TRUE 13, Endif 14, Else HW Cell is occupied 1s, If ({i] = X) then M1 Match, 16, flag = TRUE 7 Return(i) 18. Exit Ut End of the execution 19. Else 1 No match 20. i= i mop hel Hf Closed looping 2. Endif 2. Endlt 23. EndWhile 24, If (flag= FALSE) and (i=index) then #7 No match and reach to the starting point 25. Print “The table is overflow” 26, Endif 27. Endlf 28. Stop Note Step 9 in the above algorithm. Here, we assume that whenever a key value is deleted from the hash table its corresponding entries are made negative instead of NULL. Writing an algorithm for deleting a key value is straightforward and is left as an exercise. Drawback of closed hashing and its remedies The major drawback of closed hashing is that, as half of the hash table is filled, there is a tendency towards clustering: that is key values are clustered in large groups and as a result a sequential search becomes slower and slower. This kind of clustering is typically known as primary clustering. The following are some solutions known to avoid this situation: (a) Random probing, (b) Double hashing or rehashing (©) Quadratic probing. Let us briefly discuss each of the above solutions. Random probing. This method uses a pseudo random number generator to generate a random sequence of locations, rather than an ordered sequence as was the case in the linear probing method, The random sequence generated by the pseudo random number generator contains all the positions between | and h, the highest location of the hash table. An example of a pseudo random number generator that produces such a random sequence of locations is given below: i+ m) Mop h +1Tables 203 where i is a number in the sequence, and m and h are integers that are relatively prime to each other (that is, their greatest common divisor is 1). Fot example, suppose m = 5 and A= 11 and initially i = 2, then the above-mentioned pseudo random number generator generates the sequence as: 8, 3,9, 4, 10,5, 11, 6, 1, 7, 2 We stop producing the numbers when the first location is duplicated. Observe that here all the numbers between | and I are generated but randomly. We can avoid primary clustering if the probe follows the said random sequence. Double hashing. Random hashing however is not free form clustering. Another type of clustering, called secondary clustering, is involved here. In particular, clustering occurs when two keys are hashed into the same location. In such an instance, if the same sequence of locations is generated for two different keys by the random probing method then clustering takes place. An alternative approach to avoid the secondary clustering problem is to use a second hash function in addition to the first one. This second hash function results in the value of m for the pscudo random number generator as employed in the random probing method. This second function should be selected in such a way that the hash addresses generated by the two hhash functions are distinct and the second function generates a value m for the key k so that m and fe are relatively prime. Let us consider the following example. Suppose H,(k) is the initially used hash function and Hs(K) is the second one. These two functions are defined as H(A) = (k Moo fi) +1 Hk) = (k Mop (ht - 4) +1 Let f= 11 and & = 50 for an instance. Then, Fy(50) = 7 and #,(50) = 2. Therefore, H,(50) # (50), that is, H, and H, are independent and mm = 2, h = 11 are relatively prime. Hence, using i= [G+ 2) Mop 11] + 1, and initially i = 7, we have the random sequence as 10, 2, 5, 8, 11, 3, 6, 9, 1, 4, 7 Now, let us choose another key value which has the same hash address as that of 50 (that is, 7) with the first hash function H,. Let it be 28 (since H,(28) = 28 Mop 11 + 1 = 7). Then H,(28) = 28 mop 7+ 1 = 5 So using i = [(i +m) mop 11] with i = 7 and m = 5, we get the sequence: 2,8 3,9, 4, 10,5, 11, 6, 1,7 Thus, for the two key values where the hash address is the same and using rehashing, two different random sequences are generated, thereby alleviating the secondary clustering. Quadratic probing. Quadratic probing is a collision resolution method that eliminates the primary clustering problem of linear probing. For linear probing, if there is a collision at location i, then the next locations i + 1, #+ 2, i + 3, ete. are probed; but in quadratic probing, the next locations to be probed are i + 17, + 2%, i +34, etc, Mathematically, if is the size of the hash table and H(t) is the hash function then the quadratic probing searches the locations:204 Classic Data Structures H(&® + 2 mop fh for i = 1, 2, 3, ... Note that in quadratic probing the increment function is #. It also assumes the hash table as close (or circular) as in linear probing. This method, no doubt, substantially reduces primary clustering, but it does not probe all the locations in the table. Lemma 6.1 gives the information regarding the number of location that it can probe at most. Lemma 6.1 If h denotes the size of the hash table then the number of distinct positions that will be probed is (h + 1/2. Proof: Suppose that the hash address for a given key k is x. Then the ith probe will look like x+@ mod hax+[1+345+--+@i- 1] moh Qi-Isk ie. (6.2) Hence proved. Example: Suppose h =11 and the hash address of the key is x. Then the different locations with a quadratic probe are x,x + 1.x +4,x 49, x4 5,2 +3 with (11 + 1/2 = 6 probes. Drawhack of quadratic probing: For linear probing, it is not advisable to let the hash table get nearly tull because in that case we may have to search the entire table and thus performance degrades. For quadratic probing, the situation is even more drastic: there is no guarantee of finding an empty cell once more than half of the table gets full or even before that if the table size is not prime. Lemma 6.2 supports the above situation. Lemma 6.2 If quadratic probing is used and the table size is prime, then a new key value can always be inserted if the table is at least half full. Proof (By the method of contradiction): Let the table size A be an (odd) prime number greater than 3. We show that the first | h/2| alternate locations are distinct. Two of these locations are x4? Mop and x+j? mop h where 0 < i, j $ ||h/2], and x is the hash address of a key. Suppose by contradiction, these locations are the same, but i = j. Then x4? Mop k= x+j? Mop h or (? - 7) Mop h=0Tables 205 G-) x G+) mov h=0 Since h is prime, it follows that either i ~j or i + jis divisible by h. Again? #j, i,j < [h/2], so (i —f) Mov h # 0. The second option is also not possible as i, j < |h/2], their sum can never be m x h, for m= 1, 2,3... Thus, the first | /2 alternate locations are distinct. Since the element to be inserted can also be placed in the location to which it hashes, if there are no collisions, any element has [4/2] locations into which it can be placed. Hence, proved. Tn quadratic probing, it is also very crucial that the table size should be a prime. If the table size is not prime, the number of alternate locations can be severely reduced. As an example, if the table size is 16, (or a power of 2), then the only altemate locations would be at distances 1,4, 9, etc. 6.44 Open Hashing So far we have discussed the closed hashing methods of collision resolution. The closed hashing method deals with arrays as hash tables and thus we are able to refer quickly to random positions in the tables. But there are two main difficulties with this technique: First, it is very difficult to handle the situation of table overflow in a satisfactory manner. Second, the key values are haphazardly intermixed and, on the average, the majority of the keys are far from their hash locations, thus increasing the number of probes which degrades the overall performance. To resolve these problems another hashing method called open hashing (also called separate chaining, ot simply chaining) is known. The chaining method is discussed in the following paragraphs. The chaining method uses a hash table as an array of pointers: each pointer points a linked list. That is, here the hash table is an array of list headers. In Figure 6.9, a hash table of size 10 is considered. The index of the hash table varies from 0 to 9 and key values are tak imegers. The hash address for a key is decided by its last digit (means the right most digit 10 XI 72 =f ° ° a alee o}-Le =n 19 29 Figure 69 An open hashing. a\3 Seve Hsonso206 Classic Data Structures For a given key value, the hash address is calculated, It then searches the linked list pointed by the pointers at that location. If the element is found it returns the pointer to the node containing that key value else inserts the element at the end of that list. The implementation of open hashing is stated in the algorithm HashChaining as follows: Algorithm HashChaining Input: K is the item of interest. INSERT is a flag for the option of insertion. Output: If K is found in the hash table then return the pointer of the node which contains the key value K else insert X into the linked list when the INSERT flag is TRUE. Data structure: Hash table H having size HSIZE storing pointer to the single linked list structure, Steps: 1. index = HashFunetion(&) M Calculate the hash address of K 2. pir = Afindex) 1 pir is a pointer to any node in the list 3. flag = FALSE 11 flag for controlling the search 4. While (ptr # NULL) and (flag = FALSE) do 5. If (pr DATA = K) then 1 End of search 6 flag = TRUE 7 Return(ptr) & Exit Mf End of execution 9. Else 10. pir = pir.LINK M# Move to the next node i. Endif 12. EndWhile 13. If (flag = FALSE) then 14. Print “Key value does not exist” 15. If (INSERT) then 16. InsertEnd_SL(Hfindex}) IV Insert it into the table 1% Endif 18, Endlf 19. Stop A key value if it exist can be deleted from a hash table for which a procedure HashKeyDelere(...) can be written. This is left as an exercise for the reader. Advantages and disadvantages of chaining ‘There are several advantages of the chaining method. The most important advantages are stated below: 1. An overflow situation never arises, The hash table maintains lists which can contain any number of key valucs. 2. Collision resolution can be achieved very efficiently if the lists maintain an ordering of keys, so that keys can be searched quickly,Tables 207 3. Insertion and deletion become a quick and an easy task in open hashing. Deletion proceeds in exactly the same way as deletion of a node in a single linked list. 4, Finally, open hashing is best suitable in applications where the number of key values varies drastically as open hashing uses dynamic storage management policy. The only disadvantage of the chaining method is that of maintaining linked lists and extra storage space for link fields. Assignment 6.3 As an alternative to the collision resolution technique, bucket hashing can be used. In this method, the hash table is a collection of buckets and each bucket contains a few number of key values decided by the bucket size. Let us assume that all buckets are of the same size. Here, the hash function calculates an address of a bucket and then finally the key value is searched in that bucket. Figure 6.10 illustrates this concept for a hash table with buckets whose size is 3, (a) Compare bucket hashing with open hashing and closed hashing. (b) Write algorithms to search a key value, insert a key value and delete a key value in bucket hashing. Bucket #1] ——> Bucket #2 ———+ Bucket 3 ———> Bucket #4 ——> Buckot #5 ——> Bucket #6 ——> Figure 6.10 Bucket hashing. 6.4.5 Comparison of Collision Resolution Techniques We will conclude the discussion of hash tables by giving an analytical comparison of various collision resolution techniques discussed. Let us define the load factor, A, of a hash table as _ Total number of key values 3) Size of the hash table ome208 Classic Data Structures So A = 1.0 means that the number of key values is the same as the total capacity of the hash table. We also define S(4) and U(A) as ‘S(A) = average number of probes for a successful search. U(A) = average number of probes for an unsuccessful search. These two quantitics will measure the performance of collision resolution methods. Analysis of closed hashing To analyze the performance of closed hashing, let us assume the case of random probing and ignore the problem of clustering for the sake of simplicity. Let us first consider the case of unsuccessful search. It is evident that the probability that the first probe hits an occupied cell is 4, the load factor. The probability that a probe hit an empty cell is 1 ~ 2. The probability that the unsuccessful search terminates in exactly two probes is therefore A(1 — A). Arguing similarly this way, the probability that exactly k probes are made in an unsuccessful search is 2!-(1 - 2), The average number of probes for an unsuccessful search is therefore U(a= Duta -a=0-A) Sut (6.4a) eI 1 cape we hve a- Since As 1 and kd! = a U(ay=0- A) a (6.4b) ay Next, let us consider the case of a successful search, We can think of this problem through insertion of key values. Then the number of probes required will be exactly one more than the number of probes made in the unsuccessful search before inserting the item. Let us consider the case when the table is initially empty. In that state, key values are inserted one at a time, Now as the items are inserted, the load factor grows slowly from 0 to A. Thus, we can express the average number of probes in a successful search as ws sa)=5 f U(x) de (6.5) A similar calculation can be performed for closed hashing with linear probing. This is left as an assignment for the student.