International Journal of Mathematics and Statistics Invention (IJMSI)

E-ISSN: 2321 4767 P-ISSN: 2321 - 4759

www.ijmsi.org Volume 1 Issue 2 December. 2013 PP-19-24
www.ijmsi.org 19 | P a g e
Theoretical Analysis Of Novel Variants Of MTF Algorithm

Debashish Rout
1

Department Of Computer Science and Engineering,Vssut,Burla,India1

ABSTRACT : In this paper, we have proposed a novel variants of MTF algorithm, which we popularly call as
MPITF and MSITF.We have also performed theoritical study and comparative performance analysis of MPITF
and MSITF with MTF

I. INTRODUCTION
Data structure provides a way to store data in structure way efficiently, in the primary memory of
computer.Various operations such as search, insert and delete can be performed on a data structure. If the data
items are unsorted and stored in a linear list, each items can be searched by scanning the the list of items one by
one linearly. In linear search if multiple elements are searched, the total search time can be reduced by making
the data structure self-organizing. In self-organization datastructure the items can be re-organized after each
operation to reduce the time of future operations Thereby enhancing the performance.

1.1 List Accessing Problem
List Accessing problem or List Update problem is the method used in the self-organizing linear search.
In List Update problem a list (l) of records and a request sequence ( o ) are taken as inputs. When a record is
accessed from the list then the list is reconfigured to reduce the future search cost. When a record is accessed
from the list, some cost is provided for that. List accessing problem is mainly implemented by single linked list.
But it may be implemented through doubly linked list and tree also.

1.2 Cost Model
In the list accessing problem, we have two different models based on operations and list type. They are
Static list accessing model and Dynamic list accessing model. The Static list accessing model is the one in which
the number of items in the list is fixed and only the access operation can be performed. The Dynamic list
accessing model is the one in which the size of the list varies dynamically and all the three operations i.e. insert,
delete and access can be performed. In our work, we have considered only the static model of list accessing
problem and hence we consider only the access operation. As one of the key issues is to find out the optimal
access cost of elements on the list, we need a cost model which is an efficient tool to measure the access cost
incurred by the various list accessing algorithms. A number of cost models have been developed and used so far
but here we have considered only Full Cost Model (FCM) and Partial Cost Model (PCM). In Full Cost Model,
the cost of accessing an item in the i
th
position from the front of the list is i. In the Partial Cost Model the cost of
accessing an item in the i
th
position from the front of the list is (i-1) because we have to make (i-1) comparisons
before accessing the i
th
element in the list. So, the cost of accessing the first element in the list would be 1 in
FCM and 0 in PCM. We are illustrating both the models as follows. Suppose the list is 1, 2, 3 and the request
sequence is 1, 2, and 3. The costs of elements according to the various models are presented in below.









1.3 Application
List accessing algorithms are widely used in Data Compression. Other important applications of list
update algorithms are computing point maxima in computational geometry, resolving collisions in hash table and
dictionary maintenance. The List Accessing Problem is also of significant interest in the contest of self
organizing data structures.


Elements Access cost in
PCM
Access Cost in
FCM
1 0 1
2 1 2
3 2 3
Total cost 3 6
Theoretical Analysis Of Novel Variants Of
www.ijmsi.org 20 | P a g e
1.4 List Accessing Algorithms
The algorithm which efficiently recognizes the list and reduces the cost for the access is called a list
accessing algorithm. List accessing algorithms can be of two types such as online algorithm and offline
algorithm. In online algorithm, the request sequence is partially known. In offline algorithm, the request
sequence is fully known; online algorithms can be further classified into two types such as deterministic and
randomized. Deterministic algorithm is one which produces the same output always for a given request sequence
or the algorithm passes through same states for a given request sequence. Some of the popular deterministic
algorithms for the list accessing problem are Move-To-Front (MTF), Transpose (TRANS) and Frequency Count
(FC).
MTF: After accessing an item, it is moved towards the front of the list without changing the order of other items
in the list.
TRANS: After accessing an element, it is exchanged with its proceeding element.
FC: There is a counter for each item which counts the frequency of each item of the list according based on the
requests from the request sequence. The list is arranged in the non-increasing order of frequency count of items
in the list.In randomized online algorithm, while processing the request sequence, the algorithm makes some
random decision at some step. Some well known randomized algorithms are SPLIT, BIT, COMB and TIME-
STAMP.

1.5 Literature Review
List update problem was first studied by McCabe in 1965 with the concept of relocatable records in
serial files. He also introduced two list accessing algorithms Move-To-Front (MTF) and Transpose (TRANS).
Rivest has examined a class of heuristics for maintaining the sequential list in optimal order with respect to the
average time required to search for a specified element with an assumption of fixed probability of each search in
his experimental study.He has shown that MTF and Transpose heuristic are optimal within a constant factor.
Hester and Hirschberg have done a comprehensive survey of all permutation algorithms that modified the order
of linear search lists with an emphasis on average case analysis. Sleator and Tarjan in their seminar paper have
formally introduced the concept of competitive analysis for online deterministic list update algorithms such as
MTF, TRANS and FC using amortized analysis and potential function method. MTF is proved to be 2-
competitive where as FC and TRANS are not competitive. Irani proposed first randomized online list update
algorithm known as SPLIT which is 1.932-competitiv. Albers, Von-Stengel, and Werchner proposed a simple
randomized online algorithm-COMB that achieves a 1.6-competitive, the best randomized algorithm in literature
till date. Albers introduced the concept of look ahead in the list update problem and obtained improved
competitive ratio for deterministic online algorithms. Reingold and Westbrook have proposed an optimal offline
algorithm which runs in time O(2
l
l!n) where l is the length of the list and n is the length of request sequence.
Bachrach et al. have provided an extensive theoretical and experimental study of online list update algorithm in
2002. The study of locality of reference in list accessing problem was initiated by Angelopoulos in 2006, where
he proved MTF is superior to all algorithms. Relatively less work has been done on the offline algorithms for the
list accessing problem. For analyzing the performance of online algorithm by competitive analysis an optimal
offline algorithm is essential. Ambhl in 2000 proved that off-line list update is NP-hard by showing a reduction
from the Minimum Feedback Arc Set Problem. In 2004, Kevin Andrew and David Gleich showed that the
randomized BIT algorithm is 7/4-competitive using a potential function argument. They introduced the pair-wise
property and the TIMESTAMP algorithm to show that the COMB algorithm, a Combination of the BIT and
TIMESTAMP algorithms, is 8/5-competitive. In 2009, in one of the paper a survey has been done on online
algorithms for self organizing sequential search .

1.6 Our Contribution
In this paper, we have proposed a novel variants of MTF algorithm, which we popularly call as MPITF
and MSITF.We have also performed theoritical study and comparative performance analysis of MPITF and
MSITF with MTF.


1.7 Organization of paper
The paper has been introduced in section 1.MTF algorithm is discussed in section 2. Proposed
algorithms are discussed in section 3. Section 4 Theoritical analysis of the algorithms. The paper is concluded in
section 5 followed by a set of references

II. MTF ALGORITHM
In deterministic online algorithm, we mainly concentrate on MTF and FC list accessing algorithms.
Here we develop new variants of both MTF and FC algorithms and also the concept of hybrid algorithms.Here
Theoretical Analysis Of Novel Variants Of
www.ijmsi.org 21 | P a g e
we have made access cost comparisons between MTF algorithm and my proposed algorithms and try to find
some formulas for my proposed algorithms.

2.1Algorithm MTF:On accessing an element x in the list, is moved to the front of the list
2.1 Variants of MTF
2.2.1 MPITF (Move Preceding Item to Front):-On accessing an item x in the list, just preceding item of x in
the list is moved to the front of the list.
2.2.2 MSITF (Move Succeeding Item to Front of the List):- On accessing an item x in the list, just succeeding
item of x in the list is moved to the front of the list.

III. THEORITICAL ANALYSIS
Special Type of Request Sequences:-
For characterization of request sequences we have considered the following parameters.
1. Size of the list:-l
2. Size of the request sequence:n
Type 1(T
1)
: Request sequence is exactly the same as that of the list.
Type 2(T
2)
: Request sequence is the reverse order as that of the list.
Type 3(T
3)
: Request sequence is a permutation of arbitrary order as that of the list (except Type 1 and Type 2).
Type 4(T
4)
: Request sequence consist of any single element of the list at position p repeated n times where 1 p
n.
Type 5(T
5)
: Request sequences consist of more than one element each repeated at least once.

[1] For MPITF Algorithm:-(Move Preceding Item)
- No of Best case sequences =No of worst case sequences for n>=2.Where the n is the length of request
sequence.
- No of best case sequences and worst sequences=n, for n>2.
The worst case value=nl-1.where n is the size of request sequence and l is the size of the list.

[2] For MSITF Algorithm(Move Succeeding Item to the Front):-
- No. of worst case sequences=n-1
- The worst case value is nl-1.(Where n is the size of the request sequence and l is the size of the list)
Proposition 1:-For T
1
request sequence of size n, the cost of MPITF is given by




P
i
--Position of the element
Proposition 2:-For T
2
request sequence of size n, the cost of MPITF is given by









Proposition 3:-For T
1
request sequence of size n, the cost of MSITF is given by

=
=
n
i
i
P T
MPITF
C
1
1
) (

= =

=
=
n
n
i
i
n
n
i
i
n
P
P
C T
MPITF
C
2
1
2
2
) (
n,even
n, odd
Theoretical Analysis Of Novel Variants Of
www.ijmsi.org 22 | P a g e









Proposition 4:-For T
2
request sequence of size n, the cost of MPITF is given by




Proposition 5:-For Type 4 request sequence of size n, the cost of MSITF is given by







Proposed Theorems:-
Theorem 1:-For Type 2 request sequence of size n, the cost of MPITF is given by



Proof-
l=Size of the list
n=Size of the request sequence
n = o n L =
Assume that ) 2 (Type C
MSITF
=
n
P

n
P = 1 ) 1 .( + n n
Basic Case
1
P =1(1-1) +1
R. H.S=n (n-1) +1
=1(1-1) +1
=1

L. H. S-
As list contains a single element, the position of that element is always 1. The reverse order of
that single element is same as that of the single element of that list. There is no succeeding element present in the
list for that single element. From that list only one request sequence is generated containing single element. By
help of MSITF, the total accessed cost is 1.
L.H. S= R. H. S
Hence
1
P is true.
Inductive Step
Let
n
P is true for n=k i.e.

k
P = 1 ) 1 .( + k k
= 1
2
+ k k
Now we have to prove that
1 + k
P = 1 ) 1 1 .( 1 + + + k k
=
1
1 ) ( 1
2
+ + =
+ +
k k
k k

+
=

=
+
=
2
1
) (
2
1
0
1 2
2
1
2
1
n
P
P
T
MSITF
C
n
o
o
n
e
i
n,even
n, odd
1 ) 1 .( ) (
2
+ = n n T
MSITF
C
= ) (
4
T
MSITF
C
1 .......... 1 , ,
1
1
2
=

=
+ =
n i Where
C C
n C
i i
n
1 ) 1 .( ) (
2
+ = n n T
MSITF
C
Theoretical Analysis Of Novel Variants Of
www.ijmsi.org 23 | P a g e

Let the elements of the list of size k be
k
l l l ..... ,
2 1
and the elements of request sequence be
k
r r r ....
2 , 1
such that all the elements of the list are present in the request sequence. Let (k+1) th element
1 + k
l occurs after
k
l in the list and
1 + k
r occur after
k
r in the request sequence. The access cost of k element
is 1
2
+ k k . Whereas the total access cost including (k+1) th element in request sequence in the list is
=
k
P + 2k
As every time the next cost is increased with a factor of 2n i.e. 2k.
= 1
2
+ k k +2k
= 1
2
+ + k k
So L.H.S= R.H.S

Hence
1 + k
P is true.
So the expression is true for all n.
Theorem 2:-For Type 1 request sequence of size n, the cost of MPITF is given by





P
i
--Position of the element
Proof-
Let
) (n C
MPITF

cost is denoted by

=
=
n
i
i n
P C
1

Where
n
C is a proposition about natural number such that
n
C is true?
Basic Case
1
1
1
1 1
= =

= i
P C
Let there is a single element in the list l i.e.
1
l and single element in the request sequence r i.e.
1
r .
As a single element present in the list, so the position of that element is 1 always. The cost of the
request sequence by using MPITF is 1 always.
So
1
C is true.
Inductive Step
Let
n
C is true for n=k i.e.
=
=
k
i
i k
P C
1

Now we have to prove that
+
=
= +
1
1
1
k
i
i k
P C


R.H.S=

+
=
1
1
k
i
i
P =0+1+2+3+. +k+ (k+1)
= (0+1+2+3+. +k) + (k+1)
= (

=
k
i
i
P
1
) + (k+1)
Let the elements of the list of size k be
k
l l l ..... ,
2 1
and the elements of request sequence be
k
r r r ....
2 , 1
such that all the elements of the list are present in the request sequence. Let (k+1) th element
1 + k
l occurs after
k
l in the list and
1 + k
r occur after
k
r in the request sequence. The access cost of k element is

=
=
n
i
i
P T
MPITF
C
1
1
) (
Theoretical Analysis Of Novel Variants Of
www.ijmsi.org 24 | P a g e

=
k
i
i
P
1
where as the access cost of (k+1) th element of request sequence in the list is (k+1) because position of
that (k+1) th element is (k+1) .So total access cost for the (k+1) element is

=
k
i
i
P
1
+ (k+1).
So L.H.S=R.H.S
Hence
1 + k
C

is true.
So the expression is true for all n.
IV. CONCLUSION
Here we have derived some formula for special request sequence these formula are proved by Induction
principle.

V. ACKNOWLEGDEMENT:-
This work is during my M.TECH career under supervision of Rakesh Mohanty ,Lecturer ,Department of
Computer Science and Engineering ,VSSUT ,Burla.

REFERENCES:-
[1] J. McCabe, On serial files with relocatable records, Operational Research, vol. 12, pp.609-618, 1965.
[2] G. Jr. Schay, F.W. Dauer-A Probabilistic Model for a Self Organizing File System, SIAM J.of Applied Mathematics, (15) 874-
888, 1967.
[3] P.J Burvile and J.F.C. Kingman On a model for storage and search J. of Applied Probability, 10:697-701, 1973.
[4] R. Rivest-On Self Organizing Sequential Search Heuristics, Communications of the ACM, 19, 2:63-67, 1976.
[5] J.R. Bitner Heuristics that dynamically organize data structures- SIAM J. of Computing, 8(1):82-110, 1979.
[6] G.H. Gonet, J. I, Munro and H. Suwanda, Towards self organizing linear search-FOCS, 169-174, 1979.
[7] G.H. Gonet, J. I, Munro and H. Suwanda, Exegenesis of self organizing linear search, SIAM Journal of Computing, Vol. 10,
no.3, pp.613-637, 1981.
[8] D. D. Sleator and R. E. Tarjan, Amortized efficiency of list update paging rules, Commun. ACM, vol. 28 no. 2, pp.202-208,
1985
[9] J. H. Hester, D. S. Hirschberg Self organizing linear search ACM Computing Surveys, 17(3): 295-312, 1985.
[10] J.L. Bently and C. C. McGeoch, Amortized analysis of self-organizing sequential search heuristics, CACM, vol. 28, pp. 404-
411, 1985.
[11] Nick Reingold and Jeffery Westbrook, Optimum Off-line Algorithms for the List Updates Problems-Technical Report
YALEU/DCS/TR-805, Yale University, 1-24, 1990.
[12] S. Ben David, A. Borodin and R. Karp, On the Power of Randomization in Online Algorithms- In Algorithmic a, pages 379-386,
1990.
[13] Sandy Irani, Two Results on the List Update Problem-IPL, 38(6):301-306, 1990.
[14] Boris Teia, A lower bound for randomized list updates algorithms- IPL 47:59, August 1993.
[15] Nick Reingold, Jeffery Westbrook, and Daniel D. Sleator,Randomized Competitive Algorithms for the List Update Problem-
algorithmic a, 11:15-32, 1994.
[16] Susanne Albers,Improved Randomized On-Line Algorithms for the List Update Problem- SODA, 412-419, 1995.
[17] Susanne Albers, Bernhard von Stengel and Ralph Werchner.A Combined BIT and TIMESTAMP Algorithm for the List Update
Problem - IPL, 56:135139, 1995.
[18] Rajeev Motwani, Prabhakar Raghavan, Randomized Algoritms -ACM Computing Surveys, 1996.
[19] Allan Borodin and Ran EI-Yaniv, On Randomization in Online Computation-STOC, 1997.
[20] S. Albers and M. Mitzenmacher, Revisiting the COUNTER algorithms List Update-IPL, 64(2):155-160, 1997.
[21] Rakesh Mohanty,Burle Sharma and Sasmita Tripathy, Characterization of Request Sequences for List Accessing
Problem and New Theoretical Results for MTF Algorithm -International Journal of Computer Applications (0975
8887)Volume 22 No.8, May 2011.