A Decision Tree-based Approach to Dynamic Pointcut Evaluation

Transcription

1 A Decision Tree-based Approach to Dynamic Pointcut Robert Dyer and Hridesh Rajan Department of Computer Science Iowa State University October 19, 2008

2 Overview Motivation: Dynamic PCD Approach: Decision-tree based Matching Technical Contributions: Formalization of the PCD problem Algorithms using Decision-tree structures for faster matching Use of implication relationships for partial evaluation of type predicates Robert Dyer and Hridesh Rajan 2 Decision Tree-based Dynamic PCD

3 Overview Terminology PCD a A, the set of attributes o O, the set of operators v V, the set of values Robert Dyer and Hridesh Rajan 3 Decision Tree-based Dynamic PCD

4 Overview Terminology PCD a A, the set of attributes o O, the set of operators v V, the set of values pred ::= ( a, o, v ) fact ::= ( a, v ) Robert Dyer and Hridesh Rajan 4 Decision Tree-based Dynamic PCD

5 Overview Terminology PCD a A, the set of attributes o O, the set of operators v V, the set of values pred ::= ( a, o, v ) fact ::= ( a, v ) PCD ::= pred ( PCD ) pred && PCD pred PCD join point ::= fact fact && join point Robert Dyer and Hridesh Rajan 5 Decision Tree-based Dynamic PCD

6 Overview Terminology PCD A ::= {modifier, type, name} V ::= {v : v is a modifier, type or name in the program} O ::= {==,!=} Robert Dyer and Hridesh Rajan 6 Decision Tree-based Dynamic PCD

7 Overview Terminology PCD A ::= {modifier, type, name} V ::= {v : v is a modifier, type or name in the program} O ::= {==,!=} Example PCD (modifier, ==, public) && (type,!=, void) && (name, ==, "Set") Robert Dyer and Hridesh Rajan 7 Decision Tree-based Dynamic PCD

8 Overview Terminology PCD A ::= {modifier, type, name} V ::= {v : v is a modifier, type or name in the program} O ::= {==,!=} Example PCD (modifier, ==, public) && (type,!=, void) && (name, ==, "Set") Example join point (modifier, public) && (type, FElement) && (name, "Set") Robert Dyer and Hridesh Rajan 8 Decision Tree-based Dynamic PCD

9 Overview Terminology PCD Robert Dyer and Hridesh Rajan 9 Decision Tree-based Dynamic PCD

10 Overview Terminology PCD 2 ways of viewing the problem PCDEval Robert Dyer and Hridesh Rajan 10 Decision Tree-based Dynamic PCD

11 Overview Terminology PCD 2 ways of viewing the problem PCDEval Robert Dyer and Hridesh Rajan 11 Decision Tree-based Dynamic PCD

12 Overview Overview Partial of Types Algorithm overview Order predicates for efficiency Create PCD evaluation tree(s) Add predicates to decision trees Create links to parents Robert Dyer and Hridesh Rajan 12 Decision Tree-based Dynamic PCD

13 Overview Overview Partial of Types Consider the following PCD: Pred1 (Pred2&&Pred3) Robert Dyer and Hridesh Rajan 13 Decision Tree-based Dynamic PCD

14 Overview Overview Partial of Types Order predicates for efficiency Modifiers are simple to match Makes other decision-trees disjoint (smaller) Robert Dyer and Hridesh Rajan 14 Decision Tree-based Dynamic PCD

15 Overview Overview Partial of Types Robert Dyer and Hridesh Rajan 15 Decision Tree-based Dynamic PCD

16

17

18

19

20

21

22

23

24

25

26

27

28

29

30

31

32

33 Overview Overview Partial of Types Goal: Reduce size of decision-trees Idea: Partially evaluate predicates Robert Dyer and Hridesh Rajan 33 Decision Tree-based Dynamic PCD

34 Overview Overview Partial of Types Known: B C Evaluate: A B, A C Robert Dyer and Hridesh Rajan 34 Decision Tree-based Dynamic PCD

35 Overview Overview Partial of Types Known: B C Evaluate: A B, A C A B B C Robert Dyer and Hridesh Rajan 35 Decision Tree-based Dynamic PCD

36 Overview Overview Partial of Types Known: B C Evaluate: A B, A C A B B C A C Robert Dyer and Hridesh Rajan 36 Decision Tree-based Dynamic PCD

37 Overview Overview Partial of Types Known: B C Evaluate: A B, A C A B B C A C Partially Evaluate: A B Robert Dyer and Hridesh Rajan 37 Decision Tree-based Dynamic PCD

38 Overview Created implementation in Nu virtual machine Bind and Remove primitives for deploying/un-deploying advice Synthetic micro-benchmark Measures time to Bind (add to trees) and match Varies type hierarchy depth Robert Dyer and Hridesh Rajan 38 Decision Tree-based Dynamic PCD

39 Overview Robert Dyer and Hridesh Rajan 39 Decision Tree-based Dynamic PCD

40 Old matching code - ( 40µs constant)

41 Old matching code - average case 3-50x slower worst case 3-88x slower

42 Related Work Overview Efficient Matching Techniques Dynamic Residue Partial Techniques Robert Dyer and Hridesh Rajan 42 Decision Tree-based Dynamic PCD

43 Future Work Overview Example Implementation(s) Real-world s Robert Dyer and Hridesh Rajan 43 Decision Tree-based Dynamic PCD

44 Overview Motivation: Dynamic PCD PCDs arrive dynamically PCDs might be removed later Matching the whole (loaded) system against a PCD is too slow Approach: Decision-tree based Matching Order evaluations based on cost Partially evaluate wherever possible Technical Contributions: Formalization of the PCD problem Algorithms using Decision-tree structures for faster matching Use of implication relationships for partial evaluation of type predicates Robert Dyer and Hridesh Rajan 44 Decision Tree-based Dynamic PCD

45 Overview Questions? nu/ Robert Dyer and Hridesh Rajan 45 Decision Tree-based Dynamic PCD

46

47 C. run C1. run C2. run C3. run.. class C { p u b l i c s t a t i c void run ( ) { measure { Bind.. / / to methods r e t u r n i n g C1 } measure { Bind.. / / to methods r e t u r n i n g C2 } measure { Bind.. / / to methods r e t u r n i n g C3 } } measure { C1. testmethod } measure { C2. testmethod } measure { C3. testmethod } } p u b l i c C testmethod ( ) { r e t u r n NULL }