JavaOne 2013: Garbage Collection Unleashed - Demystifying the Wizardry

© 2013 IBM Corporation

Ryan A. Sciampacone – Managed Run4me Architect 24 September 2013

Garbage Collec4on Unleashed Demys4fying the Wizardry


Important Disclaimers THE INFORMATION CONTAINED IN THIS PRESENTATION IS PROVIDED FOR INFORMATIONAL PURPOSES ONLY.

WHILST EFFORTS WERE MADE TO VERIFY THE COMPLETENESS AND ACCURACY OF THE INFORMATION CONTAINED IN THIS PRESENTATION, IT IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED.

ALL PERFORMANCE DATA INCLUDED IN THIS PRESENTATION HAVE BEEN GATHERED IN A CONTROLLED ENVIRONMENT. YOUR OWN TEST RESULTS MAY VARY BASED ON HARDWARE, SOFTWARE OR INFRASTRUCTURE DIFFERENCES.

ALL DATA INCLUDED IN THIS PRESENTATION ARE MEANT TO BE USED ONLY AS A GUIDE.

IN ADDITION, THE INFORMATION CONTAINED IN THIS PRESENTATION IS BASED ON IBM’S CURRENT PRODUCT PLANS AND STRATEGY, WHICH ARE SUBJECT TO CHANGE BY IBM, WITHOUT NOTICE.

IBM AND ITS AFFILIATED COMPANIES SHALL NOT BE RESPONSIBLE FOR ANY DAMAGES ARISING OUT OF THE USE OF, OR OTHERWISE RELATED TO, THIS PRESENTATION OR ANY OTHER DOCUMENTATION.

NOTHING CONTAINED IN THIS PRESENTATION IS INTENDED TO, OR SHALL HAVE THE EFFECT OF:

-‐ CREATING ANY WARRANT OR REPRESENTATION FROM IBM, ITS AFFILIATED COMPANIES OR ITS OR THEIR SUPPLIERS AND/OR LICENSORS

2


Your Speaker: Ryan A. Sciampacone

• Run4me Architect @ IBM (JTC) •  Interpreters • Garbage Collec4on • Plaaorm interfacing / op4miza4on • Hardware exploita4on

(E) [email protected] (T) @rsciampacone

• Visit the IBM Booth #5112 to meet other IBM Developers

3


What should you take away from all this?

• Not a tuning talk • Explore some (possibly) unexpected behavior in Java

• Your plaaorm changes performance in strange ways

• Add to your “deep knowledge” toolkit • Most importantly: To be entertained

4


UP TO SPEED Let’s get everyone…

5


Compac4on

6

Thread Graphic: Wikimedia Commons http://en.m.wikipedia.org/wiki/File:Process_vs._thread.svg

? Allocate

Heap

Object

Thread


Compac4on

7


? Allocate

Heap

Object

Thread


Compac4on

8


? Allocate

Heap

Object

Thread


Compac4on

9


Allocate

Heap

Object

Thread


Genera4onal

10

Heap

Object

Thread

New Old



Genera4onal

11

Heap

Object

Thread

New Old

Allocate



Genera4onal

12

Heap

Object

Thread

New Old

Allocate



Genera4onal

13

Heap

Object

Thread

New Old

GC Thread Graphic: Wikimedia Commons http://en.m.wikipedia.org/wiki/File:Process_vs._thread.svg


Genera4onal

14

Heap

Object

Thread

New Old



Genera4onal

15

Heap

Object

Thread

New Old

Allocate



Genera4onal

16

Heap

Object

Thread

New Old

Allocate



Genera4onal

17

Heap

Object

Thread

New Old



Genera4onal

18

Heap

Object

Thread

New Old

GC

Tenure



Genera4onal

19

Heap

Object

Thread

New Old

Global GC Thread Graphic: Wikimedia Commons http://en.m.wikipedia.org/wiki/File:Process_vs._thread.svg


Stop the World

20

Duke Logo: © Oracle Recycle Symbol: Creative Commons http://commons.wikimedia.org/wiki/File:Recycle.jpg


Concurrent

21


Java Ac4vity

GC Ac4vity


Parallel

22


Java Ac4vity

GC Ac4vity

…


ALLOCATION Quick look at the mechanics of…

23


Alloca4on

24

Padlock: Public domain http://en.wikipedia.org/wiki/File:Viking_Age_lock.png

Heap

Object

Thread A

lloca

te


Alloca4on

25

Padlock: Public domain http://en.wikipedia.org/wiki/File:Viking_Age_lock.png

Heap

Object

Thread A

lloca

te


Alloca4on

26

Padlock: Public domain http://openclipart.org/detail/22179/lock-by-nicubunu

Heap

Object

Thread A

lloca

te

Private Buffer


Alloca4on

27


Heap

Object

Thread A

lloca

te


Alloca4on

28


Heap

Object

Thread A

lloca

te


FINALIZATION Gekng around to eventually talking about something you inevitably have to…

29


The meaning of “in scope”

30



31

Simple Finalizer



32

Create instance of finalizable object



33

No references to the finalizable object at this point



34

Will obj be finalized?


Example

35


Perhaps a bit more dangerous…

36


THE ORDER OF THINGS Finaliza4on

37


What can we expect as output?

38


Example

39


So what happened?

40

New Old

FinalizableObject(2)



So what happened?

41

New Old




So what happened?

42

New Old



GC


So what happened?

43

New Old



GC

Tenure


So what happened?

44

New Old




So what happened?

45

New Old




So what happened?

46

New Old




So what happened?

47

New Old



GC


So what happened?

48

New Old



GC

Finalized!


ANCHOR POINTS AND GC Finaliza4on

49


What about anchor points?

50

finalize() finalize()


What about anchor points?

51


Example

52


Why only one?

53



Why only one?

54



Why only one?

55

New Old


Why only one?

56

New Old

Allocate?


Why only one?

57

New Old

Allocate?


Why only one?

58

New Old

Allocate? Allocate?


Why only one?

59

New Old

Allocate? Allocate?

GC


ICEBERGS What lies beneath the surface…

60

© 2013 IBM Corporation 61

Iceberg: Jesus Diaz, Gizmodo


Out of Memory (as we know and love it)

62

Heap

Object

Thread

New Old


Allocate


Not all memory is “heap” memory

65

Heap

Object

Na4ve Resource



66

Heap

Object

Na4ve Resource



67

Heap

Object

Na4ve Resource


Keep track / Use facili4es available

• Plenty of process monitoring somware out there

• Consider resource limi4ng / thronling facili4es

72


SHARING MEMORY Two for me and one for you…

73


Two threads and one field

74

Object

Thread field1

field2

field<N>

…

“Same”


Two threads and one field

75

Object

Thread field1

field2

field<N>

…

field1

field2

field<N>

…

“Different”


Example

76


The CPU Cache (High level)

77

CPU: Wikipedia Commons https://en.wikipedia.org/wiki/Central_processing_unit

(0x1F89DB04)

Memory

CPU



78


(0x1F89DB04)

Memory

CPU

Cache



79


(0x1F89DB04)

Memory

CPU

Cache



80


(0x1F89DB04)

Memory

CPU

Cache Cache CPU



81


(0x1F89DB04)

Memory

CPU

Cache Cache CPU



82


(0x1F89DB04)

Memory

CPU

Cache Cache CPU


Two threads and two fields

83

Object

Thread field1

field2

field<N>

…

“Same”


Two threads and two fields

84

Object

Thread field1

field2

field<N>

…

field1

field2

field<N>

…

“Different”


Example

85


The CPU Cache Pt. 2 (High level)

86


Memory

CPU

Cache



87


Memory

CPU

Cache

Cache Line



88


Memory

CPU

Cache

Cache Line



89


Memory

CPU

Cache Cache CPU

Cache Line



90


Memory

CPU

Cache Cache CPU

Cache Line



91


Memory

CPU

Cache Cache CPU

Cache Line


FIELD ORDERING When ordering pizza you expect the crust on the bonom…

92


Subclass with fields “far apart”

93


Subclass with fields “far apart”

94


Two threads and two fields and one subclass

95

Object

Thread

field1

field2

…

pad001

pad0NN

pad100 “Second”

pad1NN

…

field3


Two threads and two fields and one subclass

96

Object

Thread

field1

field2

…

pad001

pad0NN

pad100

pad1NN

… “Third”

field3


Example

97


So what happened?

98


So what happened?

99

field1

field2

…

pad001

pad0NN

pad100

pad1NN

…

field3


So what happened?

100

field1

field2

…

pad001

pad0NN

pad100

pad1NN

…

field3


So what happened?

101

field1 field2

…

pad001

pad0NN

pad100

pad1NN

…

field3


So what happened?

102

field1 field2

…

pad001

pad0NN

pad100

pad1NN

…

field3


So what happened?

103

field1 field2

…

pad001

pad0NN

pad100

pad1NN

…

field3

Cache Line


So what happened? (Alternate version)

104

field1

…

pad001

pad0NN

pad100

pad1NN

…

field2 field3


CACHE CAPACITY Spreading things out doesn’t always reduce collisions…

105


Small data, Big problems

106

byte[]



107

Cache Line (64 Bytes)

byte[]



108


byte[]



109

4KB 4KB


byte[]



110

4KB 4KB


byte[]



111

4KB 4KB


byte[]



112

4KB 4KB


byte[]



113

4KB 4KB


byte[]



•  Access the same index from each line •  Move to the next index •  Repeat

114

4KB 4KB


byte[]


Example

115


Cache “capacity” is less than you think

116


Cache



117


Cache



118


Cache



119


Cache



120


Cache



121


Cache



122


Cache 8 Entries


HOW MEMORY WORKS Before I forget, a few more things about…

123


Non Uniform Memory Architecture (NUMA)

124

CPU: Wikipedia Commons https://en.wikipedia.org/wiki/Central_processing_unit DRAM: Wikipedia Commons http://en.wikipedia.org/wiki/Dynamic_random-access_memory



125




126




127




128




129




130



Non Uniform Memory Architecture

• JVMs do have support • Balanced GC policy •  -‐Xnuma

• Plaaorm commands for thronling / afini4zing •  Taskset • Numactl

• Virtualiza4on

131


Resource on Memory

132


PARTING REMARKS And finally…

133


? 134


Copyright and Trademarks

© IBM Corpora4on 2013. All Rights Reserved. IBM, the IBM logo, and ibm.com are trademarks or registered trademarks of Interna4onal Business Machines Corp., and registered in many jurisdic4ons worldwide.

Other product and service names might be trademarks of IBM or other companies. A current list of IBM trademarks is available on the Web – see the IBM “Copyright and trademark informa4on” page at URL: www.ibm.com/legal/copytrade.shtml

135

JavaOne 2013: Garbage Collection Unleashed - Demystifying the Wizardry

Technology

Transcript of JavaOne 2013: Garbage Collection Unleashed - Demystifying the Wizardry