Mark-Compact Sweep Garbage Collection

1
Mark-Compact Sweep Garbage Collection
Automatic Heap Resizing A Valuable Combination
2
The Problem

• Processes require different heap sizes
• Mark sweep garbage collectors have some
  disadvantages
• What combination of heap size and garbage
  collection will increase performance?

3
Problem with Heap Sizes
Small heap
Large Heap
4
Problem with Mark Sweep Garbage Collection
Garbage Live Live Garbage Live
(a) Before Collection
Live Live Live
(b) After Collection
5
The Solution

• Run the Dacapo Suite and GCBench benchmarks on
  the Jikes RVM using several different
  configurations
• MSwithHR
• MSwithoutHR
• MSCwithHR
• MSCwithoutHR
• NOGCwithHR

6
Benchmarks (Dacapo Suite)
Benchmark Description
Antlr parses one or more grammar files and generates a parser and lexical analyzer for each.
Fop takes an XSL-FO file, parses it and formats it, generating a PDF file.
Jython interprets a series of Python programs
Pmd analyzes a set of Java classes for a range of source code problems
Ps reads and interprets a PostScript file
7
Benchmarks (GCBench)

• focuses on modeling allocation requests in ways
  that are especially stressful to memory
• objects are allocated to create trees of
  differing heights and sizes.

8
Heap Resizing Algorithm
0 0 0.1 0.3 0.6 0.8 1
0 0.9 0.9 0.95 1 1 1
0.02 0.9 0.9 0.95 1 1 1
0.05 0.95 0.95 1 1 1 1
0.15 1 1 1.1 1.15 1.2 1.2
0.3 1 1 1.2 1.25 1.35 1.3
0.5 1 1 1.25 1.3 1.5 1.5
1 1 1 1.25 1.3 1.5 1.5

• Calculate the live ratio (amount of memory in
  use/current heap size)
• Calculate heap change ratio
• How much garbage collection has been done?
• Calculate heap change ratio based on live ratio,
  GC Load and function matrix
• New size old heap size heap change ratio
• If new size ! old size
• Change the heap size
• Inform VM that the heap size has changed

9
Results (GCBench)
10
Results (fop)
11
Garbage Collection Results
12
Garbage Collection Results
13
Conclusions

• The variation of mark compact sweep implemented
  in the Jikes RVM works just as well as the basic
  mark sweep algorithm, even in the face of
  compaction and when complemented with automatic
  heap resizing
• In some cases, we need to examine the application
  itself to determine which algorithm is better
  suited for our needs.

14
Conclusions

• Only using garbage collection gives worse
  performance than choosing heap resizing in
  conjunction with garbage collection in situations
  in which we do not have adequate memory to
  execute our entire program
• Only using heap resizing gives worse performance
  in situations in which the footprint of the
  application is much smaller than the maximum heap
  size. In these situations, a conjunction of heap
  resizing with automatic garbage collection gives
  the best performance

15
Conclusions

• While the amount of time spent in garbage
  collection and heap resizing varies in cases in
  which the programs footprint is much smaller
  than the maximum heap size, both amount
  approximately to the same execution times

16
Questions??