1 the compressor: concurrent, incremental and parallel compaction. haim kermany and erez petrank...

28
1 The Compressor: Concurrent, Incremental and Parallel Compaction. Haim Kermany and Erez Petrank Technion – Israel Institute of Technology

Post on 19-Dec-2015

240 views

Category:

Documents


0 download

TRANSCRIPT

1

The Compressor: Concurrent, Incremental and Parallel Compaction.

Haim Kermany and Erez Petrank

Technion – Israel Institute of Technology

2

The Compressor

The first compactor with one heap pass. Fully compacts all the objects in the heap. Preserves the order of the objects. Low space overhead. A parallel version and a concurrent version.

3

Garbage collection

Automatic memory management. User allocates objects Memory manager reclaims objects which “are not

needed anymore”. In practice: unreachable from roots.

4

Modern Platforms and Requirements

High performance and low pauses. SMP and Multicore platforms:

Use parallel collectors for highest efficiency Use concurrent collectors for short pauses.

Parallel (STW)High throughput

Concurrent &Parallel

short Pauses

t

t

5

Main Streams in GC Mark and Sweep

Trace objects. Go over the heap and reclaim the unmarked objects.

Reference Counting Keep the number of pointers to each object. When an object counter reaches zero, reclaim the object.

Copying Divide the heap into two spaces. Copy all the objects from one space to the other.

6

Compaction - Motivation

M&S and RC face the problem of fragmentation. Fragmentation – unused space between live objects

due to repeated allocation and reclaiming. Allocation efficiency decreases. May fail to allocate large objects. Cache behavior may be harmed.

Compaction – move all the live objects to one place in the heap. Best practice: keep order of objects for best locality.

7

Traditional Compaction Go over the heap and write the new location of

every object in its header (install a forwarding pointer).

Update all the pointers in the roots and the heap. Move the objects

Stack

Three Heap Passes

8

Agenda

Introduction: garbage collection, servers, compaction.

The Compressor: Basic technique

Obtain compaction with a single heap pass. The parallel version. The concurrent version.

Measurements Related Work. Conclusion

9

Compressor - Overview

Compute new locations of objects Fix root pointers Move objects + fix their pointers

Stack

One Heap Passplus one pass over the (small) mark-bits table.

10

Compute new locations

Computing new locations and saving this info succinctly: Heap partitioned to blocks (typically, 512 bytes). Start by computing and saving for each block the total

size of objects preceding that block (the offset vector).

10 2 3 4 5 6 7 8 9

1000 1100 1200 1300 1400 1500 1600 1700

0 50 90 125 200 275 325 350Offset vector

The Heap

Addresses

11

10 2 3 4 5 6 7 8 9

0 50 90 125 200 275 325 350

1000 1100 1200 1300 1400 1500 1600 1700

The Heap

Addresses

Offset vectorMarkbit vector

Computing A New Address

Assume a markbit vector which reflect the heap: First and the last bits of each object are set.

A new location of an object is computed from the markbit and the offset vectors: for object 5, at the 4th block the new location is:

1000 + 125 +50 = 1175.

12

Computing Offset Vector

Computed from the markbit vector. Does not require a heap pass

10 2 3 4 5 6 7 8 9

0 50 90 125 200 275 325 350

1000 1100 1200 1300 1400 1500 1600 1700

The Heap

Addresses

Offset vectorMarkbit vector

13

Properties

Single heap pass.Plus one pass over the markbit vector.

Small space overhead. Does not need a forwarding pointer.

Single threaded.

Stop-the-world.

Next: A parallel stop-the-world (STW) version. A concurrent version.

14

Parallelization – First Try

Had we divided the heap to two spaces… The application uses only one space. The Compressor compacts the objects from

one space (from-space) to the other (to-Space). Advantage: objects can be moved independently. Problem: space overhead.

15

Eliminating Space Overhead

Initially, to-space is not mapped to physical pages. It is a virtual address space.

For every (virtual) page in to-space: (a parallel loop) Map the virtual page to physical memory. Move the corresponding from-space objects and fix the

pointers. Unmap the relevant pages in from-space.

0 1 2 3 4 6 7 8 95 4 6 7 8 95

roots

0 2 31

0 1 2 3 4 5 6 7 8 9

16

Properties

All virtues of basic Compressor: Single heap pass, small space overhead.

Easy parallelization: each to-space page can be handled independently.

Stop-The-World.

17

What about Concurrency?

Problem: two copies appear when moving objects during application run.

Sync. problems between compaction and application.

Solution (Baker style): Application can only access moved objects (in to-space).

18

Concurrent Version Stop application Fix roots to new locations in to-space. Read-protect to-space and let application resume. When application touches a to-space page a trap is sprung. Trap handler moves relevant objects into the page and

unprotect the page.

0 1 2 3 4 6 7 8 95 4 6 7 8 95

roots

19

Implementation & Measurements

Implementation on the Jikes RVM. Compressor added to a simple modification of the Jikes

mark-sweep collector (main modification: allocation via local-caches). Compressor invoked once every 10 collections.

Benchmarks: SPECjbb, Dacapo, SPECjvm98. In the talk we concentrate on SPECjbb

Compared collectors: no compaction algorithms on the Jikes RVM. Some comparison to mark-sweep (MS) and an Appel

Generational collector (GenMS).

20

SPECjbb Throughput

7000

9000

11000

13000

15000

17000

1 2 3 4 5 6 7 8

Number of WareHouses

Thro

ugh

pu

t (o

ps)

CON

STW

GenMS

MS

CON = Concurrent Compressor,

STW = Parallel Compressor

21

SPECjbb pause time (ms)

Parallel Compaction

(Stop-The-World)

Mark and Sweep

generational Mark and Sweep

(full collections)

jbb 2-WH319229.37279.73

jbb 4-WH516287323

jbb 6-WH641315347

jbb 8-WH770372374

22

SPECjbb - Allocations per time

0

10

20

30

40

50

60

70

80

90

100

-200 0 200 400 600 800 1000 1200 1400

Time From resuming the application(ms)

Allo

ca

tio

ns

Kb

/ms

2WH

4WH

6WH

8WH

23

Dacapo - Allocations per time

0

5

10

15

20

25

30

-20 -10 0 10 20 30 40 50 60 70

Time from resuming the application (ms)

Allo

catio

n r

ate

jython

pmd

ps

xalan

24

Previous Work on Compaction

Early works: Two-finger, Lisp2, and the threaded algorithm [Jonkers and Morris] are single threaded and therefore create a large pause time.

[Flood et al. 2001] first parallel compaction algorithms. But has 3 heap passes and creates several dense areas.

[Abuaiadh et al. 2004] Parallel with two heap passes, not concurrent.

[Ossia et al. 2004] execute the pointer fix-up part concurrently.

25

Related Work

Numerous concurrent and parallel garbage collectors. Copying collectors [Cheney 70] compact objects

during the collection but require a large space overhead and do not retain objects order.

Savings in space overhead for copying collectors [Sachindran and Moss 2004]

[Bacon et al. 2003, Click et al. 2005] propose an incremental compaction. But it uses a read barrier, and does not keep the order of objects.

26

Complexity Comparisons

Heap Passes

Mark-bits Passes

Remarks

Jonkers-Morris

20Two traces of the heap are interleaved with the two passes.

SUN 200130

IBM 200421Mark-bits table is small (at most 1/32 of the heap size) The

Compressor11

27

Conclusion

The Compressor: The first compactor that passes over the

heap only once. Plus one pass over the mark-bits vector.

Fully compacts all the objects in the heap. Preserves the order of the objects. Low space overhead. Uses memory services to obtain parallelism. Uses traps to obtain concurrency.

28

Questions