Implementation of

Relational
Operations
R&G - Chapters
12 and 14

Introduction
Todays topic: QUERY PROCESSING
Some database operations are EXPENSIVE
Can greatly improve performance by being
smart
e.g., can speed up 1,000,000x over nave
approach
Main weapons are:
1. clever implementation techniques for operators
2. exploiting relational algebra equivalences
3. using statistics and cost models to choose among
these.

A Really Bad Query Optimizer

For each Select-From-Where query block
Create a plan that:
Forms the cross product
of the FROM clause
Applies the WHERE clause

predicates

Then, as needed:
tables
Apply any projections and output
expressions
Apply duplicate elimination and/or ORDER BY

Cost-based Query Sub-System

Queries

Select *
From Blah B
Where B.blah = blah

Query Parser
Query Optimizer
Plan
Generator

Plan Cost
Estimator

Query Plan Evaluator

Catalog Manager

Schem
a

Statistic
s

The Query Optimization Game

Goal is to pick a good plan
Good = low expected cost, under cost
model
Degrees of freedom:
access methods
physical operators
operator orders

Roadmap for this topic:

First: implementing individual operators
Then: optimizing multiple operators

Relational Operations
We will consider how to implement:
Selection ( ) Select a subset of rows.
Projection ( ) Remove unwanted columns.
Join ( >
<) Combine two relations.
Set-difference ( - ) Tuples in reln. 1, but not in reln. 2.
Union ( ) Tuples in reln. 1 and in reln. 2.
Q: What about Intersection?

Schema for Examples

Sailors (sid: integer, sname: string, rating: integer, age

Reserves (sid: integer, bid: integer, day: dates, rname:
Similar to old schema; rname added for variations.
Sailors:
Each tuple is 50 bytes long, 80 tuples per page, 500
pages.
[S]=500, pS=80.
Reserves:
Each tuple is 40 bytes, 100 tuples per page, 1000
pages.
[R]=1000, pR=100.

Simple Selections
R. attr op value ( R)

SELECT
FROM
WHERE

*
Reserves R
R.rname < C%

How best to perform? Depends on:

what indexes are available
expected size of result
Size of result approximated as
(size of R) * selectivity
selectivity estimated via statistics we will discuss
shortly.

Our options
If no appropriate index exists:
Must scan the whole relation
cost = [R]. For reserves = 1000 I/Os.

Our options
With index on selection attribute:
1. Use index to find qualifying data entries
2. Retrieve corresponding data records
Total cost = cost of step 1 + cost of step 2
For reserves, if selectivity = 10% (100 pages, 10000
tuples):
If clustered index, cost is a little over 100 I/Os;
If unclustered, could be up to 10000 I/Os! unless
CLUSTERED

Index entries
direct search for
data entries

Data entries

UNCLUSTERED

Data entries
(Index File)
(Data file)

Data Records

Data Records

Refinement for unclustered

indexes
1. Find qualifying data entries.
2. Sort the rids of the data records to be retrieved.
3. Fetch rids in order.
Each data page is looked at just once (though # of
such pages likely to be higher than with
clustering).
UNCLUSTERED

Data entries
(Index File)
(Data file)

Data Records

General Selection Conditions

(day<8/9/94 AND rname=Paul) OR bid=5 OR sid=3

First, convert to conjunctive normal form (CNF):

(day<8/9/94 OR bid=5 OR sid=3 ) AND
(rname=Paul OR bid=5 OR sid=3)
We only discuss the case with no ORs
Terminology:
A B-tree index matches terms that involve only
attributes in a prefix of the search key. e.g.:
Index on <a, b, c> matches a=5 AND b= 3, but not
b=3.

2 Approaches to General
Selections
Approach I:
1. Find the cheapest access path
2. retrieve tuples using it
3. Apply any remaining terms that dont match the
index

Cheapest access path: An index or file scan that

we estimate will require the fewest page I/Os.

Cheapest Access Path - Example

query: day < 8/9/94 AND bid=5 AND sid=3
some options:
B+tree index on day; check bid=5 and sid=3
afterward.
hash index on <bid, sid>; check day<8/9/94
afterward.

How about a B+tree on <rname,day>?

How about a B+tree on <day, rname>?
How about a Hash index on <day, rname>?

2 Approaches to General
Selections
Approach II: use 2 or more matching indexes.
1. From each index, get set of rids
2. Compute intersection of rid sets
3. Retrieve records for rids in intersection
4. Apply any remaining terms
EXAMPLE: day<8/9/94 AND bid=5 AND sid=3
Suppose we have an index on day, and another index on sid.
Get rids of records satisfying day<8/9/94.
Also get rids of records satisfying sid=3.
Find intersection, then retrieve records, then check bid=5.

Projection

SELECT

DISTINCT

R.sid,
R.bid

Issue is removing duplicates.

FROM

Reserves

Use sorting!!
1. Scan R, extract only the needed attributes
2. Sort the resulting set
3. Remove adjacent duplicates
Cost:
Reserves with size ratio 0.25 = 250 pages.
With 20 buffer pages can sort in 2 passes, so:
1000 +250 + 2 * 2 * 250 + 250 = 2500
I/Os

Projection -- improved
Modify the external sort algorithm:
Modify Pass 0 to eliminate unwanted fields.
Modify Passes 1+ to eliminate duplicates.
Cost:
Reserves with size ratio 0.25 = 250 pages.
With 20 buffer pages can sort in 2 passes, so:
1. Read 1000 pages
2. Write 250 (in runs of 40 pages each)
3. Read and merge runs

Total cost = 1000 + 250 +250 = 1500.

Other Projection Tricks

If an index search key contains all wanted attrs:
Do index-only scan
Apply projection techniques to data entries (much
smaller!)
If a B+Tree index search key prefix has all wanted
attrs:
Do in-order index-only scan
Compare adjacent tuples on the fly (no sorting
required!)

Query Execution Framework

SELECT DISTINCT name, gpa
FROM Students

One possible query execution plan:

name, gpa

Distinct
name, gpa

Sort
name, gpa

HeapScan

Iterators

iterator

Relational operators are all subclasses of the class

iterator:
class iterator {
void init();
tuple next();
void close();
iterator &inputs[];
// additional state goes here
}
Note:
Edges in the graph are specified by inputs (max 2, usually)
Any iterator can be input to any other!

Example: Sort

class Sort extends iterator {

void init();
tuple next();
void close();
iterator &inputs[1];
int numberOfRuns;
DiskBlock runs[];
RID nextRID[];
}

init():
generate the sorted runs on disk (passes 0 to n-1)
Allocate runs[] array and fill in with disk pointers.
Initialize numberOfRuns
Allocate nextRID array and initialize to first RID of each run
next():
nextRID array tells us where were up to in each run
find the next tuple to return based on nextRID array
advance the corresponding nextRID entry
return tuple (or EOF -- End of Fun -- if no tuples remain)
close():
deallocate the runs and nextRID arrays

Postgres Version
src/backend/executor/nodeSort.c
ExecInitSort (init)
ExecSort (next)
ExecEndSort (close)
The encapsulation stuff is hardwired into
the Postgres C code
Postgres predates even C++!
See src/backend/execProcNode.c for the code
that
dispatches the methods explicitly!

Joins

SELECT
FROM
WHERE

*
Reserves R1, Sailors S
R1.sid=S1.sid

Joins are very common.

R S is large; so, R S followed by a
selection is inefficient.
Many approaches to reduce join cost.
Join techniques we will cover today:
1. Nested-loops join
2. Index-nested loops join
3. Sort-merge join

Simple Nested Loops Join

< S: foreach tuple r in R do
R>
foreach tuple s in S do
if ri == sj then add <r, s> to
result
Cost = (pR*[R])*[S] + [R] = 100*1000*500 + 1000 IOs
At 10ms/IO, Total time: ???
What if smaller relation (S) was outer?
What assumptions are being made here?
What is cost if one relation can fit entirely in
memory?

Page-Oriented Nested Loops Join

< S:foreach page bR in R do
R>
foreach page bS in S do
foreach tuple r in bR do
foreach tuple s in bSdo
if ri == sj then add <r, s> to result
Cost = [R]*[S] + [R] = 1000*500 + 1000
If smaller relation (S) is outer, cost = 500*1000 + 500
Much better than nave per-tuple approach!

Block Nested Loops Join

Page-oriented NL doesnt exploit extra
buffers :(
Idea to use memory efficiently:
R&S

block of R tuples
(B-2 pages)

Join Result

...

...

...
Input buffer for S

Output buffer

Cost: Scan outer + (#outer blocks * scan inner)

#outer blocks =
# of pages of outer / blocksize

Examples of Block Nested Loops

Join
Say we have B = 100+2 memory buffers
Join cost = [outer] + (#outer blocks * [inner])
#outer blocks = [outer] / 100
With R as outer ([R] = 1000):
Scanning R costs 1000 IOs (done in 10 blocks)
Per block of R, we scan S; costs 10*500 I/Os
Total = 1000 + 10*500.
With S as outer ([S] = 500):
Scanning S costs 500 IOs (done in 5 blocks)
Per block of S, we can R; costs 5*1000 IOs
Total = 500 + 5*1000.

Index Nested Loops Join

< S:foreach tuple r in R do
R>
foreach tuple s in S where ri == sj do
add <r, s> to result
lookup(ri)
R:

ri

INDEX on S

Data entries

sj

S:
Data Records

Index Nested Loops Join

< S:foreach tuple r in R do
R>
foreach tuple s in S where ri == sj do
add <r, s> to result

Cost = [R] + ([R]*pR) * cost to find matching S tuples

If index uses Alt. 1, cost = cost to traverse tree from root to leaf.
For Alt. 2 or 3:
1. Cost to lookup RID(s); typically 2-4 IOs for B+Tree.
2. Cost to retrieve records from RID(s); depends on clustering.
Clustered index: 1 I/O per page of matching S tuples.
Unclustered: up to 1 I/O per matching S tuple.
What assumption is made here?

Sort-Merge Join 1.
Example:
SELECT
FROM
WHERE

sid
22
28
31
44
58

Sort R on join attr(s)

2. Sort S on join attr(s)
3. Scan sorted-R and
sorted-S in tandem, to
find matches

*
Reserves R1, Sailors S1
R1.sid=S1.sid

sname rating age

dustin
7
45.0
yuppy
9
35.0
lubber
8
55.5
guppy
5
35.0
rusty
10 35.0

sid
28
28
31
31
31
58

bid
103
103
101
102
101
103

day
12/4/96
11/3/96
10/10/96
10/12/96
10/11/96
11/12/96

rname
guppy
yuppy
dustin
lubber
lubber
dustin

Cost of Sort-Merge Join

Cost: Sort R + Sort S + ([R]+[S])
But in worst case, last term could be [R]*[S] (very unlikely!)
Q: what is worst case?
Suppose B = 35 buffer pages:
Both R and S can be sorted in 2 passes
Total join cost = 4*1000 + 4*500 + (1000 + 500) = 7500

Suppose B = 300 buffer pages:

Again, both R and S sorted in 2 passes
Total join cost = 7500

Block-Nested-Loop cost = 2500 15,0

Other Considerations
1. An important refinement:
Do the join during the final merging pass of sort !
If have enough memory, can do:
1. Read R and write out sorted runs
2. Read S and write out sorted runs
3. Merge R-runs and S-runs, and find
R
S matches
><
Cost = 3*[R] + 3*[S]
Q: how much memory is enough (will answer next time )

2. Sort-merge join an especially good choice if:

one or both inputs are already sorted on join attribute(s)
output is required to be sorted on join attributes(s)
Q: how to take these savings into account? (stay tuned )

Summary
A virtue of relational DBMSs:
queries are composed of a few basic operators
The implementation of these operators can be carefully tuned
Many alternative implementation techniques for each
operator
No universally superior technique for most operators.
Must consider available alternatives
Called Query optimization -- we will study this topic soon!