indexing and execution may 20 th, 2002. indexing - recap primary file organization: heap, sorted,...
TRANSCRIPT
Indexing and Execution
May 20th, 2002
Indexing - Recap
• Primary file organization: heap, sorted, hashed.
• Primary vs. secondary
• Clustered vs. unclustered
• Dense vs. sparse
• Composite search key indexes.
• Not all combinations are possible.
• Next: B+-trees.
Tree-Based Indexes• ``Find all students with gpa > 3.0’’
– If data is in sorted file, do binary search to find first such student, then scan to find others.
– Cost of binary search can be quite high.
• Simple idea: Create an `index’ file.
Can do binary search on (smaller) index file!
Page 1 Page 2 Page NPage 3 Data File
k2 kNk1 Index File
Tree-Based Indexes (2)
P0
K1 P
1K 2 P
2K m
P m
index entry
10* 15* 20* 27* 33* 37* 40* 46* 51* 55* 63* 97*
20 33 51 63
40 Root
B+ Tree: The Most Widely Used Index
• Insert/delete at log F N cost; keep tree height-balanced. (F = fanout, N = # leaf pages)
• Minimum 50% occupancy (except for root). Each node contains d <= m <= 2d entries. The parameter d is called the order of the tree.
Index Entries
Data Entries
Root
Example B+ Tree
• Search begins at root, and key comparisons direct it to a leaf.
• Search for 5*, 15*, all data entries >= 24* ...
17 24 30
2* 3* 5* 7* 14* 16* 19* 20* 22* 24* 27* 29* 33* 34* 38* 39*
13
B+ Trees in Practice• Typical order: 100. Typical fill-factor: 67%.
– average fanout = 133
• Typical capacities:– Height 4: 1334 = 312,900,700 records
– Height 3: 1333 = 2,352,637 records
• Can often hold top levels in buffer pool:– Level 1 = 1 page = 8 Kbytes
– Level 2 = 133 pages = 1 Mbyte
– Level 3 = 17,689 pages = 133 MBytes
Inserting a Data Entry into a B+ Tree
• Find correct leaf L.
• Put data entry onto L.– If L has enough space, done!– Else, must split L (into L and a new node L2)
• Redistribute entries evenly, copy up middle key.
• Insert index entry pointing to L2 into parent of L.
• This can happen recursively– To split index node, redistribute entries evenly, but
push up middle key. (Contrast with leaf splits.)
Insertion in a B+ Tree
Insert (K, P)• Find leaf where K belongs, insert• If no overflow (2d keys or less), halt• If overflow (2d+1 keys), split node, insert in parent:
• If leaf, keep K3 too in right node• When root splits, new root has 1 key only
K1 K2 K3 K4 K5
P0 P1 P2 P3 P4 p5
K1 K2
P0 P1 P2
K4 K5
P3 P4 p5
(K3, ) to parent
Insertion in a B+ Tree
80
20 60 100 120 140
10 15 18 20 30 40 50 60 65 80 85 90
10 15 18 20 30 40 50 60 65 80 85 90
Insert K=19
Insertion in a B+ Tree
80
20 60 100 120 140
10 15 18 19 20 30 40 50 60 65 80 85 90
10 15 18 20 30 40 50 60 65 80 85 9019
After insertion
Insertion in a B+ Tree
80
20 60 100 120 140
10 15 18 19 20 30 40 50 60 65 80 85 90
10 15 18 20 30 40 50 60 65 80 85 9019
Now insert 25
Insertion in a B+ Tree
80
20 60 100 120 140
10 15 18 19 20 25
30 40
50 60 65 80 85 90
10 15 18 20 25 30 40 60 65 80 85 9019
After insertion
50
Insertion in a B+ Tree
80
20 60 100 120 140
10 15 18 19 20 25
30 40
50 60 65 80 85 90
10 15 18 20 25 30 40 60 65 80 85 9019
But now have to split !
50
Insertion in a B+ Tree
80
20 30 60 100 120 140
10 15 18 19 20 25
60 65 80 85 90
10 15 18 20 25 30 40 60 65 80 85 9019
After the split
50
30 40
50
Deleting a Data Entry from a B+ Tree
• Start at root, find leaf L where entry belongs.• Remove the entry.
– If L is at least half-full, done! – If L has only d-1 entries,
• Try to re-distribute, borrowing from sibling (adjacent node with same parent as L).
• If re-distribution fails, merge L and sibling.
• If merge occurred, must delete entry (pointing to L or sibling) from parent of L.
• Merge could propagate to root, decreasing height.
Deletion from a B+ Tree
80
20 30 60 100 120 140
10 15 18 19 20 25
60 65 80 85 90
10 15 18 20 25 30 40 60 65 80 85 9019
Delete 30
50
30 40
50
Deletion from a B+ Tree
80
20 30 60 100 120 140
10 15 18 19 20 25
60 65 80 85 90
10 15 18 20 25 40 60 65 80 85 9019
After deleting 30
50
40 50
May change to 40, or not
Deletion from a B+ Tree
80
20 30 60 100 120 140
10 15 18 19 20 25
60 65 80 85 90
10 15 18 20 25 40 60 65 80 85 9019
Now delete 25
50
40 50
Deletion from a B+ Tree
80
20 30 60 100 120 140
10 15 18 19 20 60 65 80 85 90
10 15 18 20 40 60 65 80 85 9019
After deleting 25Need to rebalanceRotate
50
40 50
Deletion from a B+ Tree
80
19 30 60 100 120 140
10 15 18 19 20
60 65 80 85 90
10 15 18 20 40 60 65 80 85 9019
Now delete 40
50
40 50
Deletion from a B+ Tree
80
19 30 60 100 120 140
10 15 18 19 20
60 65 80 85 90
10 15 18 20 60 65 80 85 9019
After deleting 40Rotation not possibleNeed to merge nodes
50
50
Deletion from a B+ Tree
80
19 60 100 120 140
10 15 18 19 20
50
60 65 80 85 90
10 15 18 20 60 65 80 85 9019
Final tree
50
Issues in Indexing
• Multi-dimensional indexing:– how do we index regions in space?– Document collections?– Multi-dimensional sales data– How do we support nearest neighbor queries?
• Indexing is still a hot and unsolved problem!
Indexing Exercise #1
• The Purchase table:– date, buyer, seller, store, product
• Reports are generated once a day.
• What’s the best file-organization/indexing strategy?
Indexing Exercise #2
• Airline database: Reservations table --– flight#, seat#, date#, occupied, customer-id
Indexing Exercise #3
• Web log application: load all the logs every night into a database.
• Generate reports every day (for curious professors).
Query Execution
Query compiler
Execution engine
Index/record mgr.
Buffer manager
Storage manager
storage
User/Application
Queryupdate
Query executionplan
Record, indexrequests
Page commands
Read/writepages
Query Execution Plans
Purchase Person
Buyer=name
City=‘seattle’ phone>’5430000’
buyer
(Simple Nested Loops)
SELECT S.snameFROM Purchase P, Person QWHERE P.buyer=Q.name AND Q.city=‘seattle’ AND Q.phone > ‘5430000’
Query Plan:• logical tree• implementation choice at every node• scheduling of operations.
(Table scan) (Index scan)
Some operators are from relationalalgebra, and others (e.g., scan, group)are not.
The Leaves of the Plan: Scans
• Table scan: iterate through the records of the relation.
• Index scan: go to the index, from there get the records in the file (when would this be better?)
• Sorted scan: produce the relation in order. Implementation depends on relation size.
How do we combine Operations?• The iterator model. Each operation is implemented
by 3 functions:– Open: sets up the data structures and performs
initializations– GetNext: returns the the next tuple of the result.– Close: ends the operations. Cleans up the data
structures.
• Enables pipelining!• Contrast with data-driven materialize model.• Sometimes it’s the same (e.g., sorted scan).
Implementing Relational Operations
• We will consider how to implement:– Selection ( ) Selects a subset of rows from relation.– Projection ( ) Deletes unwanted columns from
relation.– Join ( ) Allows us to combine two relations.– Set-difference Tuples in reln. 1, but not in reln. 2.– Union Tuples in reln. 1 and in reln. 2.– Aggregation (SUM, MIN, etc.) and GROUP BY
Schema for Examples
• Purchase:– Each tuple is 40 bytes long, 100 tuples per page, 1000
pages (i.e., 100,000 tuples, 4MB for the entire relation).
• Person:– Each tuple is 50 bytes long, 80 tuples per page, 500
pages (i.e, 40,000 tuples, 2MB for the entire relation).
Purchase (buyer:string, seller: string, product: integer),
Person (name:string, city:string, phone: integer)
Simple Selections• Of the form• With no index, unsorted: Must essentially scan the whole
relation; cost is M (#pages in R).• With an index on selection attribute: Use index to find
qualifying data entries, then retrieve corresponding data records. (Hash index useful only for equality selections.)
• Result size estimation:
(Size of R) * reduction factor.
SELECT *FROM Person RWHERE R.phone < ‘543%’
R attr valueop R. ( )
Using an Index for Selections• Cost depends on #qualifying tuples, and clustering.
– Cost of finding qualifying data entries (typically small) plus cost of retrieving records.
– In example, assuming uniform distribution of phones, about 54% of tuples qualify (500 pages, 50000 tuples). With a clustered index, cost is little more than 500 I/Os; if unclustered, up to 50000 I/Os!
• Important refinement for unclustered indexes: 1. Find sort the rid’s of the qualifying data entries. 2. Fetch rids in order. This ensures that each data page is looked at
just once (though # of such pages likely to be higher than with clustering).
Two Approaches to General Selections
• First approach: Find the most selective access path, retrieve tuples using it, and apply any remaining terms that don’t match the index:– Most selective access path: An index or file scan that
we estimate will require the fewest page I/Os.– Consider city=“seattle AND phone<“543%” :
• A hash index on city can be used; then, phone<“543%” must be checked for each retrieved tuple.
• Similarly, a b-tree index on phone could be used; city=“seattle” must then be checked.
Intersection of Rids• Second approach
– Get sets of rids of data records using each matching index.
– Then intersect these sets of rids.– Retrieve the records and apply any remaining terms.
Implementing Projection
• Two parts: (1) remove unwanted attributes, (2) remove duplicates from the result. • Refinements to duplicate removal:
– If an index on a relation contains all wanted attributes, then we can do an index-only scan.
– If the index contains a subset of the wanted attributes, you can remove duplicates locally.
SELECT DISTINCT R.name, R.phoneFROM Person R
Equality Joins With One Join Column
• R S is a common operation. The cross product is too large. Hence, performing R S and then a selection is too inefficient.
• Assume: M pages in R, pR tuples per page, N pages in S, pS tuples per page.– In our examples, R is Person and S is Purchase.
• Cost metric: # of I/Os. We will ignore output costs.
SELECT *FROM Person R, Purchase SWHERE R.name=S.buyer
Simple Nested Loops Join
• For each tuple in the outer relation R, we scan the entire inner relation S. – Cost: M + (pR * M) * N = 1000 + 100*1000*500 I/Os: 140
hours!
• Page-oriented Nested Loops join: For each page of R, get each page of S, and write out matching pairs of tuples <r, s>, where r is in R-page and S is in S-page.– Cost: M + M*N = 1000 + 1000*500 (1.4 hours)
For each tuple r in R dofor each tuple s in S do
if ri == sj then add <r, s> to result
Index Nested Loops Join
• If there is an index on the join column of one relation (say S), can make it the inner. – Cost: M + ( (M*pR) * cost of finding matching S tuples)
• For each R tuple, cost of probing S index is about 1.2 for hash index, 2-4 for B+ tree. Cost of then finding S tuples depends on clustering.– Clustered index: 1 I/O (typical), unclustered: up to 1 I/O per
matching S tuple.
foreach tuple r in R doforeach tuple s in S where ri == sj do
add <r, s> to result
Examples of Index Nested Loops• Hash-index on name of Person (as inner):
– Scan Purchase: 1000 page I/Os, 100*1000 tuples.– For each Person tuple: 1.2 I/Os to get data entry in index, plus 1
I/O to get (the exactly one) matching Person tuple. Total: 220,000 I/Os. (36 minutes)
• Hash-index on buyer of Purchase (as inner):– Scan Person: 500 page I/Os, 80*500 tuples.– For each Person tuple: 1.2 I/Os to find index page with data
entries, plus cost of retrieving matching Purchase tuples. Assuming uniform distribution, 2.5 purchases per buyer (100,000 / 40,000). Cost of retrieving them is 1 or 2.5 I/Os depending on clustering.
Block Nested Loops Join• Use one page as an input buffer for scanning the
inner S, one page as the output buffer, and use all remaining pages to hold ``block’’ of outer R.– For each matching tuple r in R-block, s in S-page, add
<r, s> to result. Then read next R-block, scan S, etc.
. . .
. . .
R & SHash table for block of R
(k < B-1 pages)
Input buffer for S Output buffer
. . .
Join Result
Sort-Merge Join (R S)• Sort R and S on the join column, then scan them to
do a ``merge’’ on the join column.– Advance scan of R until current R-tuple >= current S
tuple, then advance scan of S until current S-tuple >= current R tuple; do this until current R tuple = current S tuple.
– At this point, all R tuples with same value and all S tuples with same value match; output <r, s> for all pairs of such tuples.
– Then resume scanning R and S.
i=j
Cost of Sort-Merge Join
• R is scanned once; each S group is scanned once per matching R tuple.
• Cost: M log M + N log N + (M+N)– The cost of scanning, M+N, could be M*N
(unlikely!)
• With 35, 100 or 300 buffer pages, both Person and Purchase can be sorted in 2 passes; total: 7500. (75 seconds).
Hash-Join• Partition both relations
using hash fn h: R tuples in partition i will only match S tuples in partition i.
Read in a partition of R, hash it using h2 (<> h!). Scan matching partition of S, search for matches.
Partitionsof R & S
Input bufferfor Si
Hash table for partitionRi (k < B-1 pages)
B main memory buffersDisk
Output buffer
Disk
Join Result
hashfnh2
h2
B main memory buffers DiskDisk
Original Relation OUTPUT
2INPUT
1
hashfunction
h B-1
Partitions
1
2
B-1
. . .
Cost of Hash-Join• In partitioning phase, read+write both relations; 2(M+N).
In matching phase, read both relations; M+N I/Os.• In our running example, this is a total of 4500 I/Os. (45
seconds!)• Sort-Merge Join vs. Hash Join:
– Given a minimum amount of memory both have a cost of 3(M+N) I/Os. Hash Join superior on this count if relation sizes differ greatly. Also, Hash Join shown to be highly parallelizable.
– Sort-Merge less sensitive to data skew; result is sorted.
How are we doing?
Nested loopjoin
140 hours 50 millionI/Os
Page nestedloop
1.4 hours 500,000 I/Os
Index nestedloop
36 minutes 220,000 I/Os
Sort-mergejoin
75 seconds 7500 I/O’s
Hash join 45 seconds 4500 I/O’s
Double Pipelined Join (Tukwila)
Hash Join Partially pipelined: no output
until inner read Asymmetric (inner vs. outer) —
optimization requires source behavior knowledge
Double Pipelined Hash Join
Outputs data immediately Symmetric — requires less
source knowledge to optimize