Derivation And Evaluation of Concurrent Collectors

1
Derivation And Evaluation of
Concurrent Collectors
Martin T. Vechev
University of Cambridge
David F.
Bacon, Perry Cheng and Dave Grove
IBM T.J. Watson Research Center
2
Outline

• Motivation and Benefits
• New Generalizations
• Abstract Algorithms
• Practical Algorithms
• Derivations
• Evaluation

3
Motivation

• Concurrent Collectors
• Difficult to Construct Correctly
• Initial Errors in Dijkstra and Steele Algorithms
• Difficult to Understand
• Difficult to Implement
• No systematic comparisons, largely folklore

4
Contributions

• Generalization Of Existing Mechanisms
• Abstract Collectors Based on Generalizations
• Precise, but inefficient
• New Algorithm
• Derived from the power of generalizations
• Experimental Evaluation Of 4 Concurrent GC

5
Benefits of Generalization
Steele
Dijkstra
Hybrid
Yuasa
6
Assumptions

• Single Collector Thread
• Multiple Mutator Threads
• Atomic Write Barrier
• Non-Moving

7
Why Is It Hard ?

• Concurrent Interleaving

B
D
C
R1
A
Time
GC marks B
8
Why Is It Hard ?
R2
B
B
D
D
C
C
R1
R1
A
A
Time
Mutator creates R2
GC marks B
9
Why Is It Hard ?
R2
R2
B
B
B
D
D
D
C
C
C
R1
R1
X
R1
A
A
A
Time
Mutator creates R2
Mutator removes R1
GC marks B
10
Why Is It Hard ?
R2
R2
R2
B
B
B
B
D
D
D
D
C
C
C
C
R1
R1
X
R1
A
A
A
A
Time
Mutator creates R2
Mutator removes R1
GC reclaims C D live WRONG!
GC marks B
11
Wavefront
E
R1
G
B
D
C
F
A
R2
12
Protection
Installation
Deletion
Remember Crossing Pointer R1
Remember R2
E
E
R1
R1
G
G
B
B
D
D
C
C
F
F
X
A
A
R2
R2
13
New generalizations

• Precise Wavefront
• Shade
• Precise Counting Of Cross Pointers
• Scanned Reference Count (S-RC)

14
Collector Progress (Shade)
E
G
B
D
C
F
R2
A
SHADE0
15
Collector Progress (Shade)
E
G
B
D
C
F
R2
A
SHADE1
16
Collector Progress (Shade)
E
G
B
D
C
F
R2
A
SHADE2
17
Collector Progress (Shade)
E
G
B
D
C
F
R2
A
SHADE3
18
Shade Observations

• Computed by Collector
• Generalization of the tri-color abstraction
• Different Granularities
• Different Objects

19
Scanned Reference Count (S-RC)
E
G
B
D
C
F
A
S-RC0
20
Scanned Reference Count (S-RC)
E
G
B
D
C
F
A
S-RC1
21
Scanned Reference Count (S-RC)
E
G
B
D
C
F
A
S-RC2
22
Scanned Reference Count (S-RC)
E
G
B
D
C
F
X
A
S-RC1
23
Scanned Reference Count (S-RC)
E
G
B
D
C
F
A
X
S-RC0
24
Scanned Reference Count (S-RC)
E
G
B
D
C
F
A
S-RC0
25
Outline

• Motivation and Benefits
• New Generalizations
• Abstract Algorithms
• Practical Algorithms
• Derivations
• Evaluation

26
Abstract Algorithms

• Utilize Shade and S-RC
• Installation-Based and Deletion-Based
• Mutator nominates candidates
• Does not mark objects

27
Concurrent System Structure
COLLECT() do mark() processNominated()
while (!finished)
MUTATE (obj, field, target) obj.field
target nominate(target)
28
Mutator Nominates (Installation)
E
G
B
D
C
F
A
S-RC0
NOMINATED OBJECT BUFFER
29
Mutator Nominates (Installation)
E
G
B
D
C
F
A
S-RC1
A
NOMINATED OBJECT BUFFER
30
Mutator Nominates (Installation)
E
G
B
D
S-RC1
C
F
A
S-RC1
C
A
NOMINATED OBJECT BUFFER
31
Mutator Nominates (Installation)
E
G
B
D
S-RC1
C
F
X
A
S-RC0
C
A
NOMINATED OBJECT BUFFER
32
After Mark (Installation)
COLLECT() do mark() processNominated() wh
ile (!finished)
E
G
B
D
S-RC1
C
F
A
S-RC0
C
A
NOMINATED OBJECT BUFFER
33
After Find (Installation)

• Collector Marks Object C

E
COLLECT() do mark() processNominated() wh
ile (!finished)
G
B
D
S-RC1
C
F
A
S-RC0
C
A
NOMINATED OBJECT BUFFER
34
Allocation

• In Installation-Based Collectors
• No difference
• In Deletion-Based Collectors
• Remembered Upon Allocation

35
Allocation
E
G
B
D
C
F
A
NOMINATED OBJECT BUFFER
36
Allocation
E
G
B
D
C
F
A
S-RC1
N
N
NOMINATED OBJECT BUFFER
37
Outline

• Motivation and Benefits
• New Generalizations
• Abstract Algorithms
• Practical Algorithms
• Derivations
• Evaluation

38
Practical Algorithms

• Stacks
• Non-Barriered Region
• Scanned Object behind wavefront
• S-RC affected
• Stack rescanning
• S-RC and Shade compression (tri-color)
• Reachability Effect

39
Compressing S-RC (sticky bit)
E
G
B
D
C
F
A
S-RC0
40
Compressing S-RC (sticky bit)
E
G
B
D
C
F
A
S-RC1
41
Compressing S-RC (sticky bit)
E
G
B
D
C
F
X
A
S-RC1
42
Compressing S-RC (sticky bit)
E
G
B
D
C
F
A
S-RC1
Object A Unreachable but kept Alive
43
Compressing Shade
E
G
B
D
C
F
A
SHADE0
44
Collector Progress (Shade)
E
G
B
D
C
F
A
SHADE3
45
Collector Progress (Shade)
E
G
B
D
S-RC1
C
F
R1
A
SHADE3 PRECISE 1
46
Collector Progress (Shade)
E
G
B
D
S-RC1
(NOT decremented)
C
F
R1
X
A
SHADE3 PRECISE 1
47
Collector Progress (Shade)
E
G
B
D
C
F
A
SHADE3 PRECISE 1
Object C Unreachable but kept Alive
48
Outline

• Motivation and Benefits
• New Generalizations
• Abstract Algorithms
• Practical Algorithms
• Derivations
• Evaluation

49
Deriving Dijkstra
INSTALLATION-BASED GC
Stack Regions
RESCANNED STACKS
Compress S-RC to sticky bit for ALL objects
INSTALLATION with 1-bit
Compress Shade to 1-bit for ALL objects
DIJKSTRA
50
Deriving Yuasa
DELETION-BASED GC
Stack Regions
RESCANNED STACKS
Compress S-RC to 0-bit for ALL objects
DELETION with 0-bits
NO S-RC needed gt NO rescanning
Deletion with NO rescanning
Compress Shade to 1-bit for ALL objects
YUASA
51
Deriving a New Collector (Hybrid)
DELETION-BASED GC
Stack Regions
RESCANNED STACKS
Compress S-RC to sticky bit for Allocated
objects Compress S-RC to 0-bit for Existing
objects
MIXED DELETION
Compress Shade to 1-bit for ALL objects
HYBRID
52
Algorithms
Steele
Dijkstra
Hybrid
Yuasa
53
Evaluation

• First Systematic Comparison Of Concurrent
  Collectors
• IBM J9 Production Virtual Machine
• J2ME Profile, microJit
• 512MB RAM, Pentium 4, 3GHz
• Comparison in terms of Execution time And Space
  Overhead
• Dijkstra, Steele, Yuasa, Hybrid
• Which benchmarks
• SpecJVM98 (-s100)
• Work-Based Incremental Scheme
• Collect 9K for every 6K allocated.

54
Maximum Space Usage
javac
mtrt
jack
geomean
db
jess
55
Execution Time
javac
mtrt
jack
geomean
db
jess
56
Summary

• Generalization Of Existing Mechanisms
• S-RC, Shade
• Abstract Collectors Based on Generalizations
• Precise, but inefficient (S-RC, Shade)
• New Concrete Algorithm
• Combines good properties of Yuasa and Dijkstra
• Suitable for Real-Time Domains
• Experimental Evaluation

57
On-Going Work

• More Transformations
• Formal Proof Of Correctness
• Transformations
• Unified Abstract Collector
• Formal Relation Between Algorithms

58
(No Transcript)
59
IF TIME PERMITS SLIDES
60
(No Transcript)
61
(No Transcript)
62
Abstract Object Layout
Barrier Reference Count
Computed By Mutator
Dont Sweep
Recorded
Marked
Computed By Collector
Shade
DATA
63
The Transitive Loss
P2
P2
P2
B
B
B
B
D
D
D
D
C
C
C
C
P1
P1
A
A
A
A
Time
Thread Working
Thread Working
Collector Working
Collector Working

• C was not seen
• Reclaims C is OK
• Reclaims D live!

- Installs P2
- Marks B as live

• Deletes P1
• C becomes
• unreachable

64
Common Structure Example

WriteBarrier(Obj, field, New) if (Phase
Tracing) Old Objfield
Remember(Old) Objfield New
WriteBarrier(Obj, field, New) if (Phase
Tracing) Remember(New)
Objfield New
DIJKSTRA
YUASA