Online%20Subpath%20Profiling - PowerPoint PPT Presentation

Title: Online%20Subpath%20Profiling


1
Online Subpath Profiling

• Yossi Matias
• David Oren
• Mooly Sagiv
• School of Computer Science
• Tel-Aviv University

2
Motivation for Profiling

• Feedback on dynamic program behavior
• The 80-20 rule
• Can be used by
• Computer Architects
• Compiler Writers
• Programmers
• Better program performance

3
Motivation for Profiling
4
Types of Profiling

• Vertex profiling
• No context, just count of instructions
• Edge profiling
• Branch-transition
• Profile-directed optimization
• Path profiling
• Multiple branch-transition
• Intra- or inter-procedural

5
Types of Profiling

• Offline
• Results are collected and then displayed
• User in the loop
• Online
• Results are collected and acted upon
• JIT compilation
• Display to user

6
Motivation for Subpath Profiling

• Programs may have hot subpaths
• which are part of cold paths

7
Challenges

• Large number of subpaths
• gt4M distinct subpaths of length 2,4,...,64k in
  JLex
• gt35M total subpaths
• Counting all subpaths is prohibitively expensive
• Memory
• Time
• non linear

8
Online Subpath Profiler

• Based on an adaptive sampling technique
• Identifies arbitrary hot subpaths
• Low memory overhead
• Low runtime overhead
• Online
• Appropriate for JIT-like compilers
• Can be adapted to different requirements

9
Outline

• Algorithm overview
• Adaptive sampling
• Issues
• The OSP algorithm
• Reference implementation
• Experimental results
• Related work
• Conclusion

10
Algorithm Overview

• Select on-the-fly a random sample of subpaths
• Count the popularity of sampled subpaths and
  obtain estimation by scaling
• Achieve high accuracy using limited memory

11
Adaptive Sampling

• Based on a hot-list algorithm by Gibbons and
  Matias (SIGMOD 1998)
• Sample elements from the input set
• Frequently occurring elements will be sampled
  more often
• Sampling probability determined at runtime,
  according to the allowed memory usage
• Tradeoff between overhead and accuracy
• Give an estimate of the samples accuracy

12
Concise Samples

• Uniform random sampling
• Maintain an ltid, countgt pair for each element
• The sample size can be much larger than the
  memory size
• For skewed input sets the gain is much larger
• Sampling is not applied at every block
• Vitters reservoir sampling

13
Concise Samples
14
Issues

• Encoding
• Generating a unique ID for paths
• Path length bias
• Longer or shorter paths?
• Path representation

15
The OSP Algorithm
void sampleBlock (BasicBlock b)
subpath.appendBlock (b) if (--length 0)
updateHotList (subpath.id)
skip chooseSkipValue () subpath new
subPath () sampling false
void enterBlock (BasicBlock b) if
(sampling) sampleBlock (b) else if
(--skip 0) length choosePathLength
() sampling true
16
The OSP Algorithm
17
OSP Algorithm Walkthrough
skip 5 sampling false
void enterBlock (BasicBlock b) if
(sampling) sampleBlock (b) else if
(--skip 0) length choosePathLength
() sampling true
Skipping Sampling
18
OSP Algorithm Walkthrough
skip 4 sampling false
Skipping Sampling
19
OSP Algorithm Walkthrough
skip 3 sampling false
Skipping Sampling
20
OSP Algorithm Walkthrough
skip 2 sampling false
Skipping Sampling
21
OSP Algorithm Walkthrough
skip 1 sampling false
Skipping Sampling
22
OSP Algorithm Walkthrough
skip 0 length 2 sampling true
void enterBlock (BasicBlock b) if
(sampling) sampleBlock (b) else if
(--skip 0) length choosePathLength
() sampling true
Skipping Sampling
23
OSP Algorithm Walkthrough
skip 0 length 1 sampling true
Skipping Sampling
24
OSP Algorithm Walkthrough
skip 4 length 0 sampling false
doA-doCommon 1
void sampleBlock (BasicBlock b)
subpath.appendBlock (b) if (--length 0)
updateHotList (subpath.id)
skip chooseSkipValue () subpath new
subPath () sampling false
Skipping Sampling
25
OSP Algorithm Walkthrough
skip 3 sampling false
doA-doCommon 1
Skipping Sampling
26
OSP Algorithm Walkthrough
skip 2 sampling false
doA-doCommon 1
Skipping Sampling
27
OSP Algorithm Walkthrough
skip 1 sampling false
doA-doCommon 1
Skipping Sampling
28
OSP Algorithm Walkthrough
skip 0 length 2 sampling true
doA-doCommon 1
Skipping Sampling
29
OSP Algorithm Walkthrough
skip 0 length 1 sampling true
doA-doCommon 1
Skipping Sampling
30
OSP Algorithm Walkthrough
skip 8 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
31
OSP Algorithm Walkthrough
skip 7 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
32
OSP Algorithm Walkthrough
skip 6 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
33
OSP Algorithm Walkthrough
skip 5 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
34
OSP Algorithm Walkthrough
skip 4 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
35
OSP Algorithm Walkthrough
skip 3 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
36
OSP Algorithm Walkthrough
skip 2 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
37
OSP Algorithm Walkthrough
skip 1 sampling false
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
38
OSP Algorithm Walkthrough
skip 0 length 2 sampling true
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
39
OSP Algorithm Walkthrough
skip 0 length 1 sampling true
doA-doCommon 1 doCommon-if2 1
Skipping Sampling
40
OSP Algorithm Walkthrough
skip 6 sampling false
doA-doCommon 1 doCommon-if2 2
Skipping Sampling
41
After 1000 Iterations
doCommon-if2 253 If1-doA 130 If2-doD
127 if1-doB 122 if2-doC 118 if1-doA-..-if2 65
42
Prototype Implementation

• Written in Java, using the Soot Framework
• Handles full Java
• Low memory overhead (50kB)
• Low sampling overhead (5-50)
• Sampling Skipping overhead (current
  implementation) 30-360
• High accuracy on tested benchmarks

43
Prototype Implementation

• Limited to paths of length 2n
• Favorable tradeoff
• Simple encoding
• Tested for practical performance
• Gives more weight to shorter paths
• Only implementation details!

44
Results Runtime Overhead
Program Full Overhead Sampling Overhead
JLex 93 47
FFT 36 16
HeapSort 204 56
MolDyn 241 32
RayTrace 361 41
javac 79 7
45
Results Memory Overhead
Program Program memory Profiler memory
JLex 169,728 43,304
FFT 107,416 39,742
HeapSort 107,400 48,960
MolDyn 111,800 40,864
RayTrace 108,106 65,800
46
Results Accuracy (FFT)
Rank in sample Accurate rank Count error
1 1 0.94
2 2 0.11
3 3 1.00
4 4 0.76
5 6 29.27
6 11 3.04
7 12 0.76
47
Results Incremental (FFT)
True Rank 6 12 18 24 30 36
1 2 6 1 2 2 1
2 3 4 2 1 1 2
3 1 2 3 3 3 4
4 4 1 8 4 4 3
5 5 5 7 5 5 5
48
Related Work

• Ball, Larus Efficient path profiling (MICRO
  1996)
• Larus Whole program paths (PLDI 1999)
• Melski, Reps Interprocedural path profiling (CC
  1999)
• Taub, Schechter, Smith Ephemeral instrumentation
  for lightweight program profiling (2000)
• Sastry, Bodik, Smith Rapid profiling via
  stratified sampling (Computer Architecture 2001)
• Bala, Duesterwald, Banerjia Dynamo a
  transparent dynamic optimization system (PLDI
  2001)

49
Related Work

• Ball-Larus path profiler (MICRO 1996) and
  extensions
• Only Acyclic paths
• Whole Program Path (Larus, PLDI 1999)
• Uses an alphabet representing acyclic paths
• Compact image of a whole program trace
• Not online

50
Related Work

• Dynamo (PLDI 2000)
• A dynamic compiler for native code
• Locates hot traces and optimizes them
• Limits places where hot traces may start
• It would be interesting to integrate OSP into
  Dynamo

51
Limitations

• Results are only an approximation
• Other methods are approximations as well
• Guaranteed confidence and accuracy as function of
  hotness
• Context not taken into account
• Robust, works for arbitrary subpaths
• Stand alone tool
• Integrate into existing tools

52
Conclusions

• We have presented a framework for online subpath
  profiling
• We have a reference implementation
• Simple
• Efficient
• Accurate

53
Motivation for Subpath Profiling

• Programs may have hot subpaths
• which are part of cold paths

54
Extensions

• Post-processing
• Reconstruct longer paths from the sampled
  subpaths
• More memory efficient path representation
• Different path length bias

55
Online Subpath Profiler

• Based on an adaptive sampling technique
• At any moment during execution can report hottest
  subpaths encountered so far
• Can easily be adapted to different requirements