1	DBMS Internals Execution and Optimization May 10th, 2004
2	Agenda Questions on phase 2 of the project Today: DBMS internals part 2 -- Query execution Query optimization Next week: Thursday, not Monday. Mostly Phil Bernstein on meta-data management.
3	Query Execution
4	Query Execution Plans
5	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.
6	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).
7	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
8	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).
9	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. More on this later.
10	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).
11	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.
12	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.
13	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.
14	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, p_R tuples per page, N pages in S, p_S tuples per page. In our examples, R is Person and S is Purchase. Cost metric: # of I/Os. We will ignore output costs.
15	Discussion How would you implement join?
16	Simple Nested Loops Join For each tuple in the outer relation R, we scan the entire inner relation S. Cost: M + (p_R * M) * N = 1000 + 1001000500 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 + MN = 1000 + 1000500 (1.4 hours)
17	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 + ( (Mp_R) 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.
18	Examples of Index Nested Loops Hash-index on name of Person (as inner): Scan Purchase: 1000 page I/Os, 1001000 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, 80500 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.
19	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.
20	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.
21	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) But really, we can do it in 3(M+N) with some trickery. 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).
22	Hash-Join Partition both relations using hash fn h: R tuples in partition i will only match S tuples in partition i.
23	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.
24	Double Pipelined Join (Tukwila) Hash Join Partially pipelined: no output until inner read Asymmetric (inner vs. outer) — optimization requires source behavior knowledge
25	Query Optimization
26	Discussion How would you build a query optimizer?
27	Query Optimization Process (simplified a bit) Parse the SQL query into a logical tree: identify distinct blocks (corresponding to nested sub-queries or views). Query rewrite phase: apply algebraic transformations to yield a cheaper plan. Merge blocks and move predicates between blocks. Optimize each block: join ordering. Complete the optimization: select scheduling (pipelining strategy).
28	Building Blocks Algebraic transformations (many and wacky). Statistical model: estimating costs and sizes. Finding the best join trees: Bottom-up (dynamic programming): System-R Newer architectures: Starburst: rewrite and then tree find Volcano: all at once, top-down.
29	Key Lessons in Optimization There are many approaches and many details to consider in query optimization Classic search/optimization problem! Not completely solved yet! Main points to take away are: Algebraic rules and their use in transformations of queries. Deciding on join ordering: System-R style (Selinger style) optimization. Estimating cost of plans and sizes of intermediate results.
30	Operations (revisited) Scan ([index], table, predicate): Either index scan or table scan. Try to push down sargable predicates. Selection (filter) Projection (always need to go to the data?) Joins: nested loop (indexed), sort-merge, hash, outer join. Grouping and aggregation (usually the last).
31	Algebraic Laws Commutative and Associative Laws R U S = S U R, R U (S U T) = (R U S) U T R ∩ S = S ∩ R, R ∩ (S ∩ T) = (R ∩ S) ∩ T R S = S R, R (S T) = (R S) T Distributive Laws R (S U T) = (R S) U (R T)
32	Algebraic Laws Laws involving selection: s _{C AND C’}(R) = s _C(s _C’(R)) = s _C(R) ∩ s _C’(R) s _{C OR C’}(R) = s _C(R) U s _C’(R) s _C(R S) = s _C(R) S When C involves only attributes of R s _C(R – S) = s _C(R) – S s _C(R U S) = s _C(R) U s _C(S) s _C(R ∩ S) = s _C(R) ∩ S
33	Algebraic Laws Example: R(A, B, C, D), S(E, F, G) s _F=3(R S) = ? s _{A=5 AND G=9}(R S) = ?
34	Algebraic Laws Laws involving projections P_M(R S) = P_N(P_P(R) P_Q(S)) Where N, P, Q are appropriate subsets of attributes of M P_M(P_N(R)) = P_M,N(R) Example R(A,B,C,D), S(E, F, G) P_A,B,G(R S) = P_?(P_?(R) P_?(S))
35	Query Rewrites: Sub-queries SELECT Emp.Name FROM Emp WHERE Emp.Age < 30 AND Emp.Dept# IN (SELECT Dept.Dept# FROM Dept WHERE Dept.Loc = “Seattle” AND Emp.Emp#=Dept.Mgr)
36	The Un-Nested Query SELECT Emp.Name FROM Emp, Dept WHERE Emp.Age < 30 AND Emp.Dept#=Dept.Dept# AND Dept.Loc = “Seattle” AND Emp.Emp#=Dept.Mgr
37	Converting Nested Queries Select distinct x.name, x.maker From product x Where x.color= “blue” AND x.price >= ALL (Select y.price From product y Where x.maker = y.maker AND y.color=“blue”)
38	Converting Nested Queries Select distinct x.name, x.maker From product x Where x.color= “blue” AND x.price < SOME (Select y.price From product y Where x.maker = y.maker AND y.color=“blue”)
39	Converting Nested Queries Select distinct x.name, x.maker From product x, product y Where x.color= “blue” AND x.maker = y.maker AND y.color=“blue” AND x.price < y.price
40	Converting Nested Queries (Select x.name, x.maker From product x Where x.color = “blue”) EXCEPT (Select x.name, x.maker From product x, product y Where x.color= “blue” AND x.maker = y.maker AND y.color=“blue” AND x.price < y.price)
41	Semi-Joins, Magic Sets You can’t always un-nest sub-queries (it’s tricky). But you can often use a semi-join to reduce the computation cost of the inner query. A magic set is a superset of the possible bindings in the result of the sub-query. Also called “sideways information passing”. Great idea; reinvented every few years on a regular basis.
42	Rewrites: Magic Sets Create View DepAvgSal AS (Select E.did, Avg(E.sal) as avgsal From Emp E Group By E.did) Select E.eid, E.sal From Emp E, Dept D, DepAvgSal V Where E.did=D.did AND D.did=V.did And E.age < 30 and D.budget > 100k And E.sal > V.avgsal
43	Rewrites: SIPs Select E.eid, E.sal From Emp E, Dept D, DepAvgSal V Where E.did=D.did AND D.did=V.did And E.age < 30 and D.budget > 100k And E.sal > V.avgsal DepAvgsal needs to be evaluated only for departments where V.did IN Select E.did From Emp E, Dept D Where E.did=D.did And E.age < 30 and D.budget > 100K
44	Supporting Views 1. Create View PartialResult as (Select E.eid, E.sal, E.did From Emp E, Dept D Where E.did=D.did And E.age < 30 and D.budget > 100K) Create View Filter AS Select DISTINCT P.did FROM PartialResult P. Create View LimitedAvgSal as (Select F.did Avg(E.Sal) as avgSal From Emp E, Filter F Where E.did=F.did Group By F.did)
45	And Finally… Transformed query: Select P.eid, P.sal From PartialResult P, LimitedAvgSal V Where P.did=V.did And P.sal > V.avgsal
46	Rewrites: Group By and Join Schema: Product (pid, unitprice,…) Sales(tid, date, store, pid, units) Trees:
47	Rewrites:Operation Introduction Schema: (pid determines cid) Category (pid, cid, details) Sales(tid, date, store, pid, amount) Trees:
48	Schema for Some Examples Reserves: Each tuple is 40 bytes long, 100 tuples per page, 1000 pages (4000 tuples) Sailors: Each tuple is 50 bytes long, 80 tuples per page, 500 pages (4000 tuples).
49	Query Rewriting: Predicate Pushdown
50	Query Rewrites: Predicate Pushdown (through grouping)
51	Query Rewrite: Predicate Movearound
52	Query Rewrite: Predicate Movearound
53	Query Rewrite: Predicate Movearound
54	Query Rewrite Summary The optimizer can use any semantically correct rule to transform one query to another. Rules try to: move constraints between blocks (because each will be optimized separately) Unnest blocks Especially important in decision support applications where queries are very complex. In a few minutes of thought, you’ll come up with your own rewrite. Some query, somewhere, will benefit from it. Theorems?
55	Cost Estimation For each plan considered, must estimate cost: Must estimate cost of each operation in plan tree. Depends on input cardinalities. Must estimate size of result for each operation in tree! Use information about the input relations. For selections and joins, assume independence of predicates. We’ll discuss the System R cost estimation approach. Very inexact, but works ok in practice. More sophisticated techniques known now.
56	Statistics and Catalogs Need information about the relations and indexes involved. Catalogs typically contain at least: # tuples (NTuples) and # pages (NPages) for each relation. # distinct key values (NKeys) and NPages for each index. Index height, low/high key values (Low/High) for each tree index. Catalogs updated periodically. Updating whenever data changes is too expensive; lots of approximation anyway, so slight inconsistency ok. More detailed information (e.g., histograms of the values in some field) are sometimes stored.
57	Size Estimation and Reduction Factors Consider a query block: Maximum # tuples in result is the product of the cardinalities of relations in the FROM clause. Reduction factor (RF) associated with each term reflects the impact of the term in reducing result size. Result cardinality = Max # tuples * product of all RF’s. Implicit assumption that terms are independent! Term col=value has RF 1/NKeys(I), given index I on col Term col1=col2 has RF 1/MAX(NKeys(I1), NKeys(I2)) Term col>value has RF (High(I)-value)/(High(I)-Low(I))
58	Histograms Key to obtaining good cost and size estimates. Come in several flavors: Equi-depth Equi-width Which is better? Compressed histograms: special treatment of frequent values.
59	Histograms Statistics on data maintained by the RDBMS Makes size estimation much more accurate (hence, cost estimations are more accurate)
60	Histograms Employee(ssn, name, salary, phone) Maintain a histogram on salary: T(Employee) = 25000, but now we know the distribution
61	Histograms Ranks(rankName, salary) Estimate the size of Employee Ranks
62	Histograms Assume: V(Employee, Salary) = 200 V(Ranks, Salary) = 250 Then T(Employee Ranks) = = S_i=1,6 T_i T_i’ / 250 = (200x8 + 800x20 + 5000x40 + 12000x80 + 6500x100 + 500x2)/250 = ….
63	Plans for Single-Relation Queries (Prep for Join ordering) Task: create a query execution plan for a single Select-project-group-by block. Key idea: consider each possible access path to the relevant tuples of the relation. Choose the cheapest one. The different operations are essentially carried out together (e.g., if an index is used for a selection, projection is done for each retrieved tuple, and the resulting tuples are pipelined into the aggregate computation).
64	Example If we have an Index on rating: (1/NKeys(I)) * NTuples(R) = (1/10) * 40000 tuples retrieved. Clustered index: (1/NKeys(I)) * (NPages(I)+NPages(R)) = (1/10) * (50+500) pages are retrieved (= 55). Unclustered index: (1/NKeys(I)) * (NPages(I)+NTuples(R)) = (1/10) * (50+40000) pages are retrieved. If we have an index on sid: Would have to retrieve all tuples/pages. With a clustered index, the cost is 50+500. Doing a file scan: we retrieve all file pages (500).
65	Determining Join Ordering R1 R2 …. Rn Join tree: A join tree represents a plan. An optimizer needs to inspect many (all ?) join trees
66	Types of Join Trees Left deep:
67	Types of Join Trees Bushy:
68	Types of Join Trees Right deep:
69	Problem Given: a query R1 R2 … Rn Assume we have a function cost() that gives us the cost of every join tree Find the best join tree for the query
70	Dynamic Programming Idea: for each subset of {R1, …, Rn}, compute the best plan for that subset In increasing order of set cardinality: Step 1: for {R1}, {R2}, …, {Rn} Step 2: for {R1,R2}, {R1,R3}, …, {Rn-1, Rn} … Step n: for {R1, …, Rn} A subset of {R1, …, Rn} is also called a subquery
71	Dynamic Programming For each subquery Q ⊆ {R1, …, Rn} compute the following: Size(Q) A best plan for Q: Plan(Q) The cost of that plan: Cost(Q)
72	Dynamic Programming Step 1: For each {Ri} do: Size({Ri}) = B(Ri) Plan({Ri}) = Ri Cost({Ri}) = (cost of scanning Ri)
73	Dynamic Programming Step i: For each Q ⊆ {R1, …, Rn} of cardinality i do: Compute Size(Q) (later…) For every pair of subqueries Q’, Q’’ s.t. Q = Q’ U Q’’ compute cost(Plan(Q’) Plan(Q’’)) Cost(Q) = the smallest such cost Plan(Q) = the corresponding plan
74	Dynamic Programming Return Plan({R1, …, Rn})
75	Dynamic Programming Summary: computes optimal plans for subqueries: Step 1: {R1}, {R2}, …, {Rn} Step 2: {R1, R2}, {R1, R3}, …, {Rn-1, Rn} … Step n: {R1, …, Rn} We used naïve size/cost estimations In practice: more realistic size/cost estimations (next) heuristics for Reducing the Search Space Restrict to left linear trees Restrict to trees “without cartesian product” need more than just one plan for each subquery: “interesting orders”