schism fragmentation tolerant real time garbage collection
play

Schism: Fragmentation-Tolerant Real-Time Garbage Collection Fil - PowerPoint PPT Presentation

Nov 18, 2023 •176 likes •1.3k views

Schism: Fragmentation-Tolerant Real-Time Garbage Collection Fil Pizlo Luke Ziarek Peta Maj * Tony Hosking * Ethan Blanton Jan Vitek * * Friday, June 11, 2010 Why another Real Time Garbage Collector? Friday, June 11, 2010

  1. Replication-based GC • See: [Nettles-O’Toole ’93], [Cheng-Blelloch ’01] • Allows concurrent defragmentation • Two spaces: one space for reads; writes “replicated” to both spaces Works best for immutable objects. • Problem: Writes not atomic ! Loss of coherence! Application Read Write Object Original Replica Copying Friday, June 11, 2010

  2. Allocate in fragments [Siebert ’99] •All objects split into small fragments. •Fragment size is typically fixed at 32 bytes. •Fragments are linked, application must follow links on object access. Friday, June 11, 2010

  3. Allocate in fragments [Siebert ’99] •All objects split into small fragments. •Fragment size is typically fixed at 32 bytes. •Fragments are linked, application must follow links on object access. Plain Object Access cost is known statically, does not vary. Most objects require only two fragments. Friday, June 11, 2010

  4. Allocate in fragments [Siebert ’99] •All objects split into small fragments. •Fragment size is typically fixed at 32 bytes. •Fragments are linked, application must follow links on object access. Plain Object Access cost is known statically, does not vary. Most objects require only two fragments. Friday, June 11, 2010

  5. Allocate in fragments [Siebert ’99] •All objects split into small fragments. •Fragment size is typically fixed at 32 bytes. •Fragments are linked, application must follow links on object access. Array Access cost is logarithmic. Array accesses will see significant slow- down! Friday, June 11, 2010

  6. Allocate in fragments [Siebert ’99] •All objects split into small fragments. •Fragment size is typically fixed at 32 bytes. •Fragments are linked, application must follow links on object access. Array Access cost is logarithmic. Array accesses will see significant slow- down! Friday, June 11, 2010

  7. Allocate in fragments [Siebert ’99] •All objects split into small fragments. •Fragment size is typically fixed at 32 bytes. •Fragments are linked, application must follow links on object access. Bad idea for large arrays. Array Access cost is logarithmic. Array accesses will see significant slow- down! Friday, June 11, 2010

  8. Synopsis •Replication-copying Collection: • great, but only for immutable objects •Fragmented Allocation: • great, unless you have large arrays Friday, June 11, 2010

  9. Synopsis •Replication-copying Collection: • great, but only for immutable objects •Fragmented Allocation: • great, unless you have large arrays Can we combine the two? Friday, June 11, 2010

  10. Idea : combine Fragmented Allocation with Replication-Copying using Arraylets Friday, June 11, 2010

  11. A new way of exploiting Arraylets Friday, June 11, 2010

  12. A new way of exploiting Arraylets Arraylet Spine Friday, June 11, 2010

  13. A new way of exploiting Arraylets Arraylet Spine Fragments have fixed size - no external fragmentation Friday, June 11, 2010

  14. A new way of exploiting Arraylets The Arraylet Spine has variable size, which can lead to fragmentation! Arraylet Spine Fragments have fixed size - no external fragmentation Friday, June 11, 2010

  15. A new way of exploiting Arraylets But the spine is immutable ... Arraylet Spine Fragments have fixed size - no external fragmentation Friday, June 11, 2010

  16. A new way of exploiting Arraylets But the spine is immutable ... ... and replication is ideal for immutable objects Arraylet Spine Fragments have fixed size - no external fragmentation Friday, June 11, 2010

  17. Schism = arraylets + replication + fragments •Combination: •Concurrent mark-sweep GC for fixed-size fragments • Replication copying for variable-size arraylet spines •No external fragmentation for either fragments or spines •Heap access is O(1), wait-free, and coherent. Friday, June 11, 2010

  18. Friday, June 11, 2010

  19. Concurrent Replication Heap for Spines To-space for Array From-space for Array Spines Spines Concurrent Mark-Sweep Heap for Fragments Friday, June 11, 2010

  20. Concurrent Replication Heap for Spines To-space for Array From-space for Array Spines Spines Small Object Concurrent Mark-Sweep Heap for Fragments Friday, June 11, 2010

  21. Concurrent Replication Heap for Spines To-space for Array From-space for Array Spines Spines Large Array? Small Object Concurrent Mark-Sweep Heap for Fragments Friday, June 11, 2010

  22. Concurrent Replication Heap for Spines To-space for Array From-space for Array Spines Spines Large Array? Small Object Concurrent Mark-Sweep Heap for Fragments Friday, June 11, 2010

  23. Concurrent Replication Heap for Spines To-space for Array From-space for Array Spines Spines Large Array? Small Object Concurrent Mark-Sweep Heap for Fragments Friday, June 11, 2010

  24. Concurrent Replication Heap for Spines From-space for Array To-space for Array Spines Spines Large Array? Small Object Concurrent Mark-Sweep Heap for Fragments Friday, June 11, 2010

  25. Friday, June 11, 2010

  26. related work - or - how to make a complete RTGC Friday, June 11, 2010

  27. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Friday, June 11, 2010

  28. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Friday, June 11, 2010

  29. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Friday, June 11, 2010

  30. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Henrikkson ’98 Friday, June 11, 2010

  31. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Henrikkson ’98 Kalibera et al ’09 Friday, June 11, 2010

  32. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Blackburn & McKinley ’08 Henrikkson ’98 Kalibera et al ’09 Friday, June 11, 2010

  33. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Blackburn & McKinley ’08 Doligez, Leroy, Gonthier ’93, ’94 Henrikkson ’98 Kalibera et al ’09 Friday, June 11, 2010

  34. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Blackburn & McKinley ’08 Doligez, Leroy, Gonthier ’93, ’94 Puffitsch & Schoeberl ’08 Henrikkson ’98 Kalibera et al ’09 Friday, June 11, 2010

  35. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Blackburn & McKinley ’08 Doligez, Leroy, Fiji CMR* Gonthier ’93, ’94 Puffitsch & Schoeberl ’08 Henrikkson ’98 Kalibera et al * concurrent mark- ’09 region Friday, June 11, 2010

  36. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism Blackburn & McKinley ’08 S CHISM / CMR Doligez, Leroy, Fiji CMR* Gonthier ’93, ’94 Puffitsch & Schoeberl ’08 Henrikkson ’98 Kalibera et al * concurrent mark- ’09 region Friday, June 11, 2010

  37. related work Cheng & Blelloch ’01 - or - how to make a complete RTGC Siebert ’99 Schism good throughput } on-the-fly Blackburn & concurrent McKinley ’08 S CHISM / CMR time/space bounds Doligez, Leroy, Fiji CMR* Gonthier ’93, ’94 Puffitsch & Schoeberl ’08 Henrikkson ’98 Kalibera et al * concurrent mark- ’09 region Friday, June 11, 2010

  38. Tunable throughput-predictability trade-off. Friday, June 11, 2010

  39. Tunable throughput-predictability trade-off. • Schism A : completely deterministic : •arrays allocated fragmented • Schism C : optimize throughput: •allocate contiguously if possible • Schism CW : simulate worst-case execution of Schism C: •poison all fast-paths (array accesses, write barriers, allocations) Friday, June 11, 2010

  40. (very short) Summary of Results •Goal: as fast as Metronome •Goal: fragmentation tolerant like Java RTS •Goal: deterministic Friday, June 11, 2010

  41. (very short) Summary of Results •Goal: as fast as Metronome •Goal: fragmentation tolerant like Java RTS •Goal: deterministic Friday, June 11, 2010

  42. SPECjvm98 throughput summary 70% 60% (100% = HotSpot) 50% Throughput 40% 30% 20% 10% 0% Java RTS Metronome Schism Friday, June 11, 2010

  43. (very short) Summary of Results •Goal: as fast as Metronome •Goal: fragmentation tolerant like Java RTS •Goal: deterministic Friday, June 11, 2010

  44. (very short) Summary of Results ✓ •Goal: as fast as Metronome •Goal: fragmentation tolerant like Java RTS •Goal: deterministic Friday, June 11, 2010

  45. Fragger Results Friday, June 11, 2010

  46. Fragger Results Friday, June 11, 2010

  47. Fragger Results Friday, June 11, 2010

  48. Fragger Results •Amount of free memory successfully allocated under fragmentation: • HotSpot : ~ 100% • Java RTS : ~ 80% • Metronome : ~ 1% , unless using >10KB objects • Schism : ~ 100% (all objects) Friday, June 11, 2010

  49. (very short) Summary of Results ✓ •Goal: as fast as Metronome •Goal: fragmentation tolerant like Java RTS •Goal: deterministic Friday, June 11, 2010

  50. (very short) Summary of Results ✓ •Goal: as fast as Metronome ✓ •Goal: fragmentation tolerant like Java RTS •Goal: deterministic Friday, June 11, 2010

  51. Schism predictability: RTEMS* on 40MHz LEON3 Friday, June 11, 2010

  52. Schism predictability: RTEMS* on 40MHz LEON3 * Real Time Executive for Missile Systems Friday, June 11, 2010

  53. Schism predictability: RTEMS* on 40MHz LEON3 The OS/hardware platform used for NASA & ESA space missions. * Real Time Executive for Missile Systems Friday, June 11, 2010

  54. Performance baseline: C code. Friday, June 11, 2010

  55. Performance baseline: C code. Using both C and Java implementations of the CDx real-time air traffic collision detection benchmark [Kalibera et al ’09]. Friday, June 11, 2010

  56. Java (CMR, Schism) versus C on CDx real-time benchmark 120 100 Milliseconds 80 60 40 Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  57. Java (CMR, Schism) versus C on CDx real-time benchmark 120 100 Milliseconds 80 60 Min 40 Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  58. Java (CMR, Schism) versus C on CDx real-time benchmark 120 100 Milliseconds Max 80 60 Min 40 Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  59. Java (CMR, Schism) versus C on CDx real-time benchmark 120 100 Milliseconds Max 80 60 CDx performance varies between Min events due to varying number of 40 predicted collisions. Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  60. Java (CMR, Schism) versus C on CDx real-time benchmark 120 100 Milliseconds 80 70.5 60 40 Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  61. Java (CMR, Schism) versus C on CDx real-time benchmark 120 96.6 100 Milliseconds 80 70.5 60 40 Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  62. Java (CMR, Schism) versus C on CDx real-time benchmark 120 97.2 96.6 100 Milliseconds 80 70.5 60 40 Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

  63. Java (CMR, Schism) versus C on CDx real-time benchmark 112.5 120 97.2 96.6 100 Milliseconds 80 70.5 60 Schism CW refines the worst-case of 40 Schism C by accounting for GC Java Java Java Java C code Fiji CMR Schism C Schism CW Schism A Friday, June 11, 2010

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend

Garbage Collection Last time Compiling Object-Oriented Languages Today

Garbage Collection Last time Compiling Object-Oriented Languages Today

Garbage Collection Last time Compiling Object-Oriented Languages Today Motivation behind garbage collection Garbage collection basics Garbage collection performance Specific example of using GC in C++

400 views • 15 slides
Hierarchical Real Time Garbage Collection Filip Pizlo , Antony Hosking, Jan Vitek (Purdue &

Hierarchical Real Time Garbage Collection Filip Pizlo , Antony Hosking, Jan Vitek (Purdue &

Hierarchical Real Time Garbage Collection Filip Pizlo , Antony Hosking, Jan Vitek (Purdue & MSR, Purdue, Purdue & TJ Watson) LCTES 2007 San Diego, CA Real Time Java (RTJ) is a growing technology for developing robust,

762 views • 64 slides
FOX Garbage Collection FOX / GC Example 1 ex1: garbage at end let x = (1, 2) , y = let tmp =

FOX Garbage Collection FOX / GC Example 1 ex1: garbage at end let x = (1, 2) , y = let tmp =

FOX Garbage Collection FOX / GC Example 1 ex1: garbage at end let x = (1, 2) , y = let tmp = (10, 20) in tmp[0] + tmp[1] , p0 = x[0] + y , p1 = x[1] + y in (p0, p1) ebp 0x00 0x04 0x08 0x0c 0x10 0x14 0x18 0x1c 0x20 0x24 esi

1.36k views • 122 slides
GC for interactive and real-time systems Interactive or real-time app concerns Reducing length

GC for interactive and real-time systems Interactive or real-time app concerns Reducing length

GC for interactive and real-time systems Interactive or real-time app concerns Reducing length of garbage collection pause Demands guarantees for worst case performance Generational GC works if: Young generation is relatively small

377 views • 19 slides
GC for interactive and real-time systems Interactive or real-time app concerns Reducing length

GC for interactive and real-time systems Interactive or real-time app concerns Reducing length

GC for interactive and real-time systems Interactive or real-time app concerns Reducing length of garbage collection pause Demands guarantees for worst case performance Generational GC works if: Young generation is relatively small

452 views • 17 slides
Memory Management What to do when coalescing fails 5H. Memory Compaction garbage collection

Memory Management What to do when coalescing fails 5H. Memory Compaction garbage collection

4/15/2018 Memory Management What to do when coalescing fails 5H. Memory Compaction garbage collection is just another way to free 6A. Swapping to secondary storage doesnt greatly help or hurt fragmentation 5E. Dynamic Relocation

346 views • 10 slides
Garbage Collection Akim Demaille, Etienne Renault, Roland Levillain June 4, 2019 TYLA Garbage

Garbage Collection Akim Demaille, Etienne Renault, Roland Levillain June 4, 2019 TYLA Garbage

Garbage Collection Akim Demaille, Etienne Renault, Roland Levillain June 4, 2019 TYLA Garbage Collection June 4, 2019 1 / 35 Table of contents Motivations and Definitions 1 Reference Counting Garbage Collection 2 Mark and Sweep Garbage

1.21k views • 95 slides
The Basics Major concerns Explicit memory management Reclaiming objects at the right time

The Basics Major concerns Explicit memory management Reclaiming objects at the right time

The Basics Major concerns Explicit memory management Reclaiming objects at the right time Garbage collection Discriminating live objects from garbage Both Fast allocation Fast reclamation Low fragmentation 2

466 views • 19 slides
Uniprocessor Garbage Collection Techniques Presented by: Shiri Dori Shai Erera Outline

Uniprocessor Garbage Collection Techniques Presented by: Shiri Dori Shai Erera Outline

Uniprocessor Garbage Collection Techniques Presented by: Shiri Dori Shai Erera Outline What is Garbage Collection Basic Garbage Collection Techniques Advanced Techniques Incremental Garbage Collection Generational Garbage

660 views • 61 slides
Garbage Collection Jan Midtgaard Michael I. Schwartzbach Aarhus University The Garbage

Garbage Collection Jan Midtgaard Michael I. Schwartzbach Aarhus University The Garbage

Compilation 2010 Garbage Collection Jan Midtgaard Michael I. Schwartzbach Aarhus University The Garbage Collector A garbage collector is part of the runtime system It reclaims heap-allocated records (objects) that are no longer in use

959 views • 56 slides
Generational Generational Garbage Collection Garbage Collection Mirko Jerrentrup,

Generational Generational Garbage Collection Garbage Collection Mirko Jerrentrup,

Generational Generational Garbage Collection Garbage Collection Mirko Jerrentrup, Jerrentrup@interactive-software.de Overview Overview motivation at a glance issues problems and limitations conclusion Motivation

635 views • 42 slides
Time Matters Minimizing Garbage Collection Overhead with Minimal Effort Gnther Blaschek

Time Matters Minimizing Garbage Collection Overhead with Minimal Effort Gnther Blaschek

Time Matters Minimizing Garbage Collection Overhead with Minimal Effort Gnther Blaschek Philipp Lengauer 2015-11-05 Configuration Hell OpenJDK 8 681 VM parameters (1300+ including tracing and debugging flags) ~ 10 205 configurations

329 views • 14 slides
Garbage Collection for a calculus with indices Luis Francisco Ziliani Departamento de

Garbage Collection for a calculus with indices Luis Francisco Ziliani Departamento de

Motivation Different approachs for implementing -calculus Adding garbage collection to Summary Garbage Collection for a calculus with indices Luis Francisco Ziliani Departamento de Computacin Facultad de Ciencias Exactas y Naturales

1.12k views • 84 slides
Improving Flow Analyses via CFA Abstract Garbage Collection and Counting Matthew Might Olin

Improving Flow Analyses via CFA Abstract Garbage Collection and Counting Matthew Might Olin

Improving Flow Analyses via CFA Abstract Garbage Collection and Counting Matthew Might Olin Shivers Georgia Institute of Technology Northeastern University ICFP 2006 1 The Big Idea The Process 1. Add garbage collection to a

1.09k views • 83 slides
Fault-Tolerant Data Collection in Fault-Tolerant Data Collection in Heterogeneous Intelligent

Fault-Tolerant Data Collection in Fault-Tolerant Data Collection in Heterogeneous Intelligent

Fault-Tolerant Data Collection in Fault-Tolerant Data Collection in Heterogeneous Intelligent Monitoring Networks Networks Heterogeneous Intelligent Monitoring Jing Deng Department of Computer Science University of North Carolina at

335 views • 18 slides
GARBAGE BAGE CO COLLECTIO LLECTION: N: @EvaAndreasson, @Cloudera AGENDA Garbage

GARBAGE BAGE CO COLLECTIO LLECTION: N: @EvaAndreasson, @Cloudera AGENDA Garbage

GARBAGE BAGE CO COLLECTIO LLECTION: N: @EvaAndreasson, @Cloudera AGENDA Garbage Collection (101) The Good The Bad The Ugly The Challenge! 2 GARBAGE COLLECTION "When you have to shoot, shoot - don't talk!" 3

372 views • 34 slides
Decidable fragments of first order logic R. Ramanujam The Institute of Mathematical Sciences,

Decidable fragments of first order logic R. Ramanujam The Institute of Mathematical Sciences,

Decidable fragments of first order logic R. Ramanujam The Institute of Mathematical Sciences, Chennai, India jam@imsc.res.in Summary Modal logics have decent algorithmic properties, useful for specification and verification. Vardi,

460 views • 30 slides
Combinations of Theories for Decidable Fragments of First-order Logic Pascal Fontaine Loria,

Combinations of Theories for Decidable Fragments of First-order Logic Pascal Fontaine Loria,

Combinations of Theories for Decidable Fragments of First-order Logic Pascal Fontaine Loria, INRIA, Universit de Nancy (France) Montreal August 2, 2009 Montreal, August 2, 2009 1 / 15 Introduction Context / Motivation

325 views • 19 slides
Reflection calculus and conservativity spectra Lev D. Beklemishev Steklov Mathematical Institute

Reflection calculus and conservativity spectra Lev D. Beklemishev Steklov Mathematical Institute

Reflection calculus and conservativity spectra Lev D. Beklemishev Steklov Mathematical Institute of RAS, Moscow Topological methods in Logic VI Tbilisi, July 17, 2018 Research reported in this work is supported by the Russian Science

738 views • 40 slides
The Berkeley File System The Original File System Background Why is the bandwidth low? The

The Berkeley File System The Original File System Background Why is the bandwidth low? The

The Berkeley File System The Original File System Background Why is the bandwidth low? The original UNIX file system was implemented on a The file system used a 512 byte block size. This block PDP-11. size is to small with 10 ms disk

328 views • 8 slides
Undecidability of propositional separation logic and its neighbours James Brotherston Computer

Undecidability of propositional separation logic and its neighbours James Brotherston Computer

Undecidability of propositional separation logic and its neighbours James Brotherston Computer Science Seminar Institute of Cybernetics, Tallinn University of Technology 17 Nov 2011 1/ 27 Outline 1. An overview of propositional separation

948 views • 76 slides
1 A Nachos Filesystem A Nachos FilesystemOn Disk On Disk A Typical Unix File Tree A Typical

1 A Nachos Filesystem A Nachos FilesystemOn Disk On Disk A Typical Unix File Tree A Typical

Representing Files On Disk: Nachos Representing Files On Disk: Nachos An OpenFile represents a file in OpenFile(sector) active use, with a seek pointer and OpenFile Seek(offset) read/write primitives for arbitrary byte ranges. Read(char*

213 views • 5 slides
A Worst-Case Opmal Mul-Round Algorithm for Parallel Computaon of Conjuncve Queries

A Worst-Case Opmal Mul-Round Algorithm for Parallel Computaon of Conjuncve Queries

A Worst-Case Opmal Mul-Round Algorithm for Parallel Computaon of Conjuncve Queries Bas Ketsman & Dan Suciu 1 Topic of the Talk How to compute mul-joins (over graphs) ... ( x , y , z ) R ( x , y ) , S ( y , z ) , T ( z , x

454 views • 31 slides
Remote Procedure Call Client Server R e Blocked q u e s t Outline Protocol Stack

Remote Procedure Call Client Server R e Blocked q u e s t Outline Protocol Stack

RPC Timeline Remote Procedure Call Client Server R e Blocked q u e s t Outline Protocol Stack Blocked Computing Presentation Formatting y p l e R Blocked Spring 2005 CS 461 1 Spring 2005 CS 461 2 RCP Components Bulk

498 views • 7 slides

More recommend