Query Processing and Optimisation

Introduction to Databases
Query Processing and Optimisation
Prof. Beat Signer
Department of Computer Science

Vrije Universiteit Brussel
http://www.beatsigner.com
2 December 2005
Context of Today's Lecture
Programmers Users DB Admins
Application Database
Queries
Programs Schema
DBMS
DML Query DDL
Preprocessor Compiler Compiler
Program Authorisation Catalogue

Object Code Control Manager
Integrity Command Query

Checker Processor Optimiser
Transaction
Scheduler
Manager
Data
Manager Buffer Recovery
Manager Manager
Database
Manager
Access File
Methods Manager
System Data, Indices and

Based on 'Components of a DBMS', Database Systems,
Buffers System Catalogue
T. Connolly and C. Begg, Addison-Wesley 2010
May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 2

Basic Query Processing Steps
Query
Query Query Plan
Compiler
Query Catalogue
Optimiser Manager
statistics
Command Query System Catalogue

Query Result
Processor Execution Plan
Data Manager
Database

Basic Query Processing Steps ...
 Query parsing and translation (query compiler)
 check the syntax (e.g. SQL for relational DBMS)
 verify that the mentioned relations do exist and replace views
 transform the SQL query to a query plan represented by a
relational algebra expression (for relational DBMS)
- different possible relational algebra expressions for a single query
 Query optimisation (query optimiser)

 transform the initial query plan into the best possible query plan
based on the given data set
- specify the execution of single query plan operations (evaluation primitives)
• e.g. which algorithms and indices to be used
- the query execution plan is defined by a sequence of evaluation primitives
 Query evaluation (command processor)

 execute the query execution plan and return the result
Query Expression and Execution
SELECT name, street
FROM Customer, Order
WHERE Order.customerID = Customer.customerID AND status = 'open';
 Transform the SQL query to the following query plan

pname, street(sstatus="open"(order ⋈ customer))
query pname, street query pname, street

expression tree execution tree pipeline
sstatus="open" (evaluation plan) ⋈ use block nested-loop join
use status s
⋈ bitmap index status="open"
order customer order customer

note that we will later see how to optimise the query expression tree

Query Costs
 The query costs are defined by the time to answer
a query (process the query execution plan)
 Different factors contribute to the query costs
 disk access time, CPU time or even network communication time
 The costs are often dominated by the disk access time
 seek time (tS) (~4 ms)
 transfer time (tT) (e.g. 0.1 ms per disk block)
- write operations are normally slower than read operations
 For simplicity, we will use the number of block transfers

and the number of seeks as cost measure
 real systems may also take CPU costs into account

Query Costs ...
 We often compute the worst case costs where the
main memory buffer can hold only a few blocks
 we further assume that data has to be initially read from disk and
is not yet in the buffer from a previous operation

Selection Operation
 The lowest-level query processing operator for
accessing data is the file scan
 search and retrieve records for a given selection condition
 In the following we discuss different file scan algorithms
 we assume that blocks of the file are stored continously
 Linear search
 given a file with n blocks, we scan each block and check if any
records satisfy the condition
 a selection on a candidate key attribute (unique) can be
terminated after a record has been found
- average costs: tS + n/2 * tT , worst case costs: tS + n * tT
 applicable to any file regardless of ordering, the availability of
indices or the type of selection operation
Selection Operation ...
 Binary search
 an equality selection condition on a file that is ordered on the
selection attribute (n blocks) can be realised via a binary search
 note that this only works if we assume that the blocks of the file
are stored continously!
 worst case costs: log2(n) * (tS + tT)

Index-based Selection Operation
 A search algorithm that makes use of an index is called an
index scan and the index structure is called access path
 Primary index and equality on candidate key
 retrieve a single record based on the index
 costs for a B+-tree with height h: (h + 1) * (tS + tT)
 Primary index and equality on non-candidate key
 multiple records might fulfil the condition (possibly spread
over n successive blocks)
 costs for a B+-tree with height h: h * (tS + tT) + tS + n * tT
 Secondary index and equality on candidate key
 retrieve a single record based on the index
 costs for a B+-tree with height h: (h + 1) * (tS + tT)

Index-based Selection Operation ...
 Secondary index and equality on non-candidate key
 each matching record may be in a different block (matching
records spread over n blocks)
 costs for a B+-tree with height h: (h + n) * (tS + tT)
 for large number of blocks n with matching records, this can be
very expensive and cost even more than a linear scan!
 Primary index and comparison on attribute A
 we assume that the relation is sorted on attribute A
 sA  v (r)
- use index to find the first record that has a value of A  v and do a sequential
file scan from there
 sA  v (r)
- sequential file scan until A  v without using any index

Index-based Selection Operation ...
 Secondary index and comparison on attribute A
 sA  v (r) or sA  v (r)
 for a B+-tree index we can scan the leaf index blocks from the
smallest value to v or from v to the largest value
 each record may be in a different block (spread over n blocks)
 for large number of records n, this can be very expensive and cost
even more than a linear scan!

Conjunctive Selection Operation
 A conjunctive selection has the form sq1  q2  ...  qn (r)
 Conjunctive selection using a single index
 check if there is an access path available for an attribute in one of
the simple conditions qi
 use one of the approaches described before (with minimal cost) to
retrieve the records and check the other conditions in memory
 Conjunctive selection using a composite index
 use the appropriate multi-key index if available
 Conjunctive selection using multiple indices
 requires indices with record pointers
 retrieve record pointers from different indices and perform an
intersection of the sets of record pointers
- additional conditions (without index) might be checked in memory
Disjunctive Selection Operation
 A disjunctive selection has the form sq1  q2  ...  qn (r)
 Disjunctive selection using indices
 indices can only be used if there is an index for all conditions;
otherwise a linear scan of the relation has to be performed
anyway

Sorting
 Sorting in database systems is important for two reasons
 a query may specify that the output should be sorted
 the processing of some relational query operations can be
implemented more efficiently based on sorted relations
- e.g. join operation
 For relations that fit into memory, techniques like

quicksort can be used
 For relations that do not fit into memory an external
merge sort algorithm can be used

External Merge Sort Example
e 2
o 8 e 17 a 3
e 2 l 9 b 20
x 0 m 5 d 12
m 5 R0 o 8 e 2
l 9 o 7 e 17
o 7 t 12 f 21
e 17 x 0 f 3
t 12 g 5
r 25 a 3 i 11
a 3 b 20 l 9
s 19 i 11 m 5
11 blocks
r 8
create n 29 merge n 29
t 13 runs R1 r 25 o 8
n 29 r 8 o 7
i 11 s 19 r 25
b 20 t 13 r 8
x 3 s 19
d 12 d 12 t 12
f 21 f 21 t 13
f 3 f 3 w 4
w 4 R2 g 5 x 0
g 5 w 4 x 3
x 3
initial relation runs (4 blocks each) sorted output
assumption in this example: memory can hold at most M = 4 blocks

External Merge Sort
 Let us assume that there is space for M memory blocks
(1) Create runs
 repeatedly read M blocks of the initial relation, sort them and write
them back as run Ri (resulting in a total of N runs)
(2) Merge the runs (N-way merge), for N < M
 use N memory blocks to buffer the input runs (one block per run)
and one block as an output buffer
 repeat the following steps until all input buffer blocks are empty
- select the smallest record rs from all input runs and write it to the output block
• if the output block is full then write it to the disk
- remove the record rs from the buffered block of run Ri
• if the buffered block of run Ri becomes empty, then fetch the next block of the
input run Ri into the buffer

External Merge Sort ...
 If N  M then multiple merge passes are required
 in each pass continuous groups of M - 1 runs are merged
 each pass reduces the number of runs by a factor M - 1
 Cost analysis
 initial number of runs: B/M
 for a file with B blocks we need logM - 1(B/M) merge passes
 creation of the initial runs requires a read and write of each block
- 2B block transfers
 each pass reads every block and writes it to the disk
- 2B block transfers per run
- last run forms an exception since the blocks do not have to be written to disk
 the total number of block transfers for an external merge sort is
therefore B * (2 * logM-1(B/M) + 1)

External Merge Sort Example
e 2
l 9 a 3
o 8 m 5 e 2 a 3
e 2 R0 o 8 e 17 b 20
x 0 o 7 l 9 d 12
m 5 x 5 m 5 e 2
l 9 o 8 e 17
o 7 a 3 o 7 e 12
e 17 e 17 r 25 f 21
t 12 create R r 25 r 8 f 3
r 25 1 r 8 s 19 g 5
a 3 runs s 19 t 12 i 11
s 19 t 12 x 5 l 9
r 8
merge merge m 5
12 blocks
t 13 b 20 pass-1 b 20 pass-2 n 29
n 29 d 12 d 12 o 8
i 11 i 11 e 12 o 7
b 20 R2 n 29 f 21 p 1
x 3 t 13 f 3 r 25
d 12 x 3 g 5 r 8
f 21 i 11 s 19
f 3 e 12 n 29 t 12
w 4 f 21 p 1 t 13
g 5 f 3 t 13 w 1
p 1 R3 g 5 w 1 x 5
e 12 p 1 x 3 x 3
w 4
initial relation runs (3 blocks each) runs sorted output
assumption in this example: memory can hold at most M = 3 blocks

Join Operation
 Different algorithms for implementing join operations
 nested-loop join
 block nested-loop join
 index nested-loop join
 merge join
 hash join
 The query optimiser may choose an algorithm based on
cost estimates
 In the join algorithm examples, we will use the two
relations Customer and Order with the following data
 Customer has 5000 records and 100 blocks
 Order has 10 000 records and 300 blocks

Nested-Loop Join
for each tuple tr in r {
for each tuple ts in s {
if (tr and ts satisfy the join condition q) {
add tuple tr  ts to the result set
}
}
}
 A nested-loop join with the outer relation r and the inner

relation s can be used to compute a theta join r ⋈q s
 The nested-loop join algorithm requires no indices and
can be used for any join condition
 A nested-loop join is expensive since every pair of tuples
in the two relations has to be examined

Nested-Loop Join ...
 Let us assume that r has br blocks and nr tuples
and s has bs blocks and ns tuples
 In the worst case, the buffer can only hold one block of
each relation r and s
 nr * bs + br block transfers and nr + br seeks
- e.g. Customer in outer relation: 5000 * 300 + 100 = 1 500 100 block transfers and
5000 + 100 = 5100 seeks
- e.g. Order in outer relation: 10 000 * 100 + 300 = 1 000 300 block transfers and
10 000 + 300 = 10 300 seeks
 In the best case, both relations fit into memory

 br + bs block transfers and 2 seeks
 If at least one relation fits into memory, that relation
should be made the inner relation
Block Nested-Loop Join
for each block Br of r {
for each Block Bs of s {
for each tuple tr in Br {
for each tuple ts in Bs {
if (tr and ts satisfy the join condition q) {
add tuple tr  ts to the result set
}
}
}
}
}
 Variant of the nested-loop join where every block of the

inner relation is paired with every block of the outer
relation

Block Nested-Loop Join ...
 Much better worst case performance than nested-loop
join
 br * bs + br block transfers and 2 * br seeks
- e.g. Customer in outer relation: 100 * 300 + 100 = 30 100 block transfers and 200
seeks
 Other optimisations
 if the join attributes in a natural join or an equi-join form a
candidate key on the inner relation, the inner loop can terminate
on the first match
 scan inner loop alternately forward and backward
- buffered data from previous scan can be reused

Indexed Nested-Loop Join
 In an indexed nested-loop join we use an index on
the inner loop's join attribute for equi-joins/natural joins
 index lookups instead of file scans
 for each tuple tr of the outer relation r, the index is used to lookup
tuples in the inner relation s
 index might even be constructed just to compute the join
 Worst case performance
 buffer has space for one block of the outer relation r and we need
an index lookup on s for each tuple in r
- cost: br * (tS + tT) + nr * c, where c is the cost for a single selection on s
 e.g. 30-ary B+-tree index on Order relation
- tree height not greater than log15(10 000) = 4
- cost: 100 * (tS + tT) + 5000 * (4+1) (tS + tT) = 25 100 * (tS + tT)

Other Join Implementations
 Merge join
 sort both relations on the join attribute
 merge the sorted relations to join them
 Hybrid merge join
 one relation is sorted and there exists a secondary B+-tree index
on the join attribute for the second relation
- merge the sorted relation with the leaf address entries of the B+-tree
- sort the result set on the addresses of the unsorted relation's tuples
- scan the unsorted relation to fetch the data and replace the pointers
 Hash join
 uses a hash function to partition the tuples of the relations r and s
based on the join attributes
 details about hash and merge join can be found in the book

Duplicate Elimination
 Duplicates can be eliminated via sorting or hashing
 when sorting, duplicates will be placed next to each other and all
but one instance of the duplicate tuples can be removed
 duplicates can be eliminated in the different steps of an external
merge sort
- while the runs are generated
- during the intermediate merge passes
 hashing places duplicate tuples into the same bucket
 The elimination of duplicates has high costs and
therefore SQL does not eliminate duplicates by default
 has to be explicitly specified via the DISTINCT keyword
 A projection can be implemented by performing the
projection on each tuple and eliminating duplicates
Set Operations
 The union (), intersection () and set difference (-)
operators can be implemented based on a variant of
merge join after sorting or a variant of the hash join
 Hash implementation
 partition the relations r and s by using a single hash function h
which results in the hash buckets Hri and Hsi
 rs
- build an in-memory index of Hri and add the tuples of Hsi not yet present
- add the tuples in the hash index to the result
 rs
- build an in-memory index of Hri and for each tuple in Hsi probe the index and
add the tuple to the result only if it is present in the hash index
 similar implementation for difference r - s
- remove tuple from the index of Hri if present in Hsi
Expressions
 The individual relational operations that have been
discussed so far normally form part of more complex
expressions
 There are two approaches how a query execution tree
can be evaluated
 materialisation
- compute the result of an evaluation primitive and materialise (store) the new
relation on the disk
 pipelining
- pass on tuples to parent operations even while an operation is still being
executed

Materialisation
 Evaluate one operation after another starting at
the leave nodes of the query expression tree
 materialise intermediate results in temporary relations and use
those for evaluating operations at the next level
pname, street 1. compute order ⋈ customer and store relation
sstatus="open" 2. compute sstatus="open" on materialised relation and store
⋈ 3. compute pname, street on materialised relation
order customer
 A materialised evaluation is always possible

 costs of reading and writing temporary relations can be quite high
 double buffering with two output buffers for each operation
Pipelining
 Pipelining evaluates multiple operations
simultaneously by passing results of one operation
to the next one without storing the tuples on the disk
 Much cheaper than materialisation since no
I/O operations for temporary relations
 Pipelining is not always possible
 e.g. does not work for input for sorting algorithms
 Pipelines can be executed in a demand driven or in a
producer driven manner
 Demand driven or lazy pipelining (pull pipelining)
 top level operation repeatedly requests the next tuple from its
children
Pipelining ...
 Producer driven or eager pipelining (push pipelining)
 the child operators produce tuples eagerly and pass them to their
parents via a buffer
 if the buffer is full, the child operator has to wait until the parent
operator consumed some tuples
 The use of pipelining may have an impact on the types of
algorithms that can be used for a specific operation
 e.g. join with a pipelined left-hand-side input
- the left relation is never available all at once for processing
- i.e. merge join cannot be used if the inputs are not sorted
- however, we can for example use an indexed nested-loop join

Query Optimisation
 There are alternative ways for evaluating a given query
 different equivalent expressions (query expression trees)
 different potential algorithms for each operation of the expression
pname, street pname, street

sstatus="open" ⋈
⋈ sstatus="open"
order customer order customer

Query Optimisation
 There can be enormous differences in terms of
performance between different query evaluation plans for
the same query
 e.g. seconds vs. days to execute the same query
 Cost-based query optimisation
(1) generate logically equivalent expressions by using a set of
equivalence rules
(2) annotate the expressions to get alternative query evaluation
plans (e.g. which algorithms to be used)
(3) select the cheapest plan based on the estimated costs
 Estimation of query evaluation costs based on
 statistical information from the catalogue manager in combination
with the expected performance of the algorithms
Equivalence Rules
 Conjunctive selection operations can be deconstructed
into a sequence of individual selections
 sq1  q2 (E) = sq1 (sq2(E))
 Selection operations are commutative
 sq1 (sq2(E)) = sq2 (sq1(E))
 Cascade of projection operations (only final one)
 pA1 (pA2(...(pAn(E))...)) = pA1(E)
 Selections can be combined with cartesian products and
theta joins
 sq (E1  E2) = E1 ⋈q E2
 sq1 (E1 ⋈q2 E2) = E1 ⋈q1q2 E2

Equivalence Rules ...
 Theta join (and natural join) operations are commutative
 E1 ⋈q E2 = E2 ⋈q E1
 note that the order of attributes is ignored
 Natural join operations are associative
 (E1 ⋈ E2) ⋈ E3 = E1 ⋈ (E2 ⋈ E3)
 Theta joins are associative in the following manner
 (E1 ⋈q1 E2) ⋈q2q3 E3 = E1 ⋈q1q3 (E2 ⋈q2 E3)
 where q2 contains attributes only from E2 and E3
 Union and intersection operations are commutative
 E1  E2 = E2  E1
 E1  E2 = E2  E1

Equivalence Rules ...
 Union and intersection operations are associative
 (E1  E2)  E3 = E1  (E2  E3)
 (E1  E2)  E3 = E1  (E2  E3)
 The selection operation distributes over union,
intersection and set difference
 sP (E1 - E2) = sP (E1) - sP (E2)
 The projection distributes over the union operation
 pA (E1  E2) = (pA (E1))  (pA (E1))
 Note that this is only a selection of equivalence rules

Heuristic Optimisation
 Cost-based optimisation can be expensive
 a DBMS may use some heuristics to reduce the number of
cost-based choices
 A heuristic optimisation transforms the query expression
tree by using a set of rules that typically improve the
execution performance
 perform selection as early as possible
- reduces the number of tuples
 perform projection as early as possible
- reduces the number of attributes
 perform most restrictive selection and join operations (smallest
result size) before other operations

Homework
 Study the following chapters of the
Database System Concepts book
 chapter 12
- sections 12.1-12.8
- Query Processing
 chapter 13
- sections 13.1-13.7
- Query Optimization

References
 A. Silberschatz, H. Korth and S. Sudarshan,
Database System Concepts (Sixth Edition),
McGraw-Hill, 2010

Next Lecture
Transaction Management
2 December 2005

Query Processing and Optimisation - Intr

Uploaded by

Copyright:

Available Formats

Query Processing and Optimisation - Intr

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Query Processing and Optimisation - Intr

Uploaded by

Copyright:

Available Formats

Introduction to Databases

Prof. Beat Signer

Department of Computer Science

Program Authorisation Catalogue

Integrity Command Query

System Data, Indices and

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 2

Command Query System Catalogue

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 3

 Query optimisation (query optimiser)

 Query evaluation (command processor)

 Transform the SQL query to the following query plan

query pname, street query pname, street

order customer order customer

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 5

 For simplicity, we will use the number of block transfers

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 6

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 7

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 9

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 10

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 11

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 12

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 14

 For relations that fit into memory, techniques like

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 15

initial relation runs (4 blocks each) sorted output

assumption in this example: memory can hold at most M = 4 blocks

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 16

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 17

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 18

initial relation runs (3 blocks each) runs sorted output

assumption in this example: memory can hold at most M = 3 blocks

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 19

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 20

 A nested-loop join with the outer relation r and the inner

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 21

 In the best case, both relations fit into memory

 Variant of the nested-loop join where every block of the

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 23

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 24

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 25

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 26

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 29

⋈ 3. compute pname, street on materialised relation

 A materialised evaluation is always possible

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 32

pname, street pname, street

order customer order customer

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 33

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 35

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 36

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 37

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 38

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 39

May 3, 2019 Beat Signer - Department of Computer Science - [email protected] 40

You might also like