1 / 29

Performance Pathologies in Hardware Transactional Memory

Explore the performance pathologies in hardware transactional memory systems and propose enhancements to improve conflict resolution.

augustinew
Download Presentation

Performance Pathologies in Hardware Transactional Memory

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Performance Pathologies in Hardware Transactional Memory Jayaram Bobba, Kevin E. Moore, Haris Volos, Luke Yen, Mark D. Hill, Michael M. Swift, and David A. Wood Multifacet Project (www.cs.wisc.edu/multifacet) Computer Sciences Dept., University of Wisconsin Madison

  2. Executive Summary (1/2) • Multiple Hardware Transactional Memory Designs • Best Performance ? • Depends ! Livelock University of Wisconsin Madison

  3. pa●thol●o●gy: any deviation from a normal, healthy or efficient condition Executive Summary (2/2) • Why? • Performance Pathologies • What to do? • Better Conflict Resolution Livelock Performance PathologiesPerformance University of Wisconsin Madison

  4. Talk Outline • Motivation • Base HTM Results • Performance Pathologies • Enhanced HTM Systems University of Wisconsin Madison

  5. Motivation • Which HTM system performs best? • Different Assumptions • Hardware • Broadcast vs Directory Based • Software: • Continuous Transactions vs Critical Section Transactions • How to compare? University of Wisconsin Madison

  6. HTM Design Space Stanford TCC Illinois Bulk No HTMs (yet) LL Herlihy/Moss TM MIT LTM Intel/Brown VTM MIT UTM Wisconsin LogTM EE EL Conflict Resolution: Adhoc “Better By Design” ? University of Wisconsin Madison

  7. Experimental Setup • Base HTM Systems • Common Hardware Platform • 32 core CMP • Directory-Based Coherence • Common TM Programming Model • Critical Section Based Transactions • Common Workloads • 5 SPLASH + 2 Microbenchmarks (Btree, LFUCache) LL EL EE University of Wisconsin Madison

  8. And the winner is… • Low Contention: Similar • High Contention: Depends Performance Pathologies Livelock University of Wisconsin Madison

  9. Talk Outline • Motivation • Base HTM Results • Performance Pathologies • Enhanced HTM Systems University of Wisconsin Madison

  10. LL EL EE Performance Pathologies • StarvingElder • SerializedCommit • RestartConvoy • FriendlyFire • DuelingUpgrades • FutileStall • StarvingWriter University of Wisconsin Madison

  11. System LL • Conflict Detection: Lazy Detect conflicts at commit time (Validation) • Version Management: Lazy New value elsewhere, Update on commit + Abort - Commit • Conflict Resolution: Committer Wins compare with TCC, Bulk University of Wisconsin Madison

  12. T1 T2 T3 load A store A ABORT load A store A ABORT load A store A ABORT COMMITING STALLED ABORTING StarvingElder • CAUSE • Committer Wins • EFFECT(S) • Starvation • Load Imbalance • FIX • Elder gets Priority ? Time University of Wisconsin Madison

  13. load A load A load A load A store A ABORT ABORT ABORT load A load A load A load A store A ABORT ABORT ABORT load A load A load A RestartConvoy • CAUSE • Fast Abort + • Immediate Restart • EFFECT(S) • Resource Contention • Serialized Execution • ≈ Lock Convoys • FIX • Stagger Restarts ? T1 T2 T4 T3 University of Wisconsin Madison

  14. Pathologies LL • Pathology (% ExecutionTime) • Enhancement • Linear Backoff on Abort • Eliminate RestartConvoys LL LL_B University of Wisconsin Madison

  15. Enhanced LL ( ) LL_B • Pathology (% ExecutionTime) • Performance LL_B LL University of Wisconsin Madison

  16. Performance Pathologies • StarvingElder • SerializedCommit • RestartConvoy • FriendlyFire • DuelingUpgrades • FutileStall • StarvingWriter LL EL EE University of Wisconsin Madison

  17. System EL • Conflict Detection: Eager • Detect conflicts immediately • Version Management: Lazy • New Value elsewhere, Update on commit + Abort - Commit • Conflict Resolution: • Requester Wins • Exponential Backoff on Abort compare with HMTM, LTM University of Wisconsin Madison

  18. T1 T2 T3 load A store A ABORT load A ABORT store A load A ABORT store A ABORT load A ABORT store A FriendlyFire • CAUSE • Requester Wins • EFFECT(S) • Unnecessary Aborts • Livelock • FIX • Priority Based Conflict Resolution ? University of Wisconsin Madison

  19. Pathologies EL • Pathology (% ExecutionTime) • Enhancement • Timestamp Based Conflict Resolution • Eliminate FriendlyFire EL EL_T University of Wisconsin Madison

  20. Enhanced EL ( ) EL_T • Pathology (% ExecutionTime) • Performance EL EL_T University of Wisconsin Madison

  21. Performance Pathologies • StarvingElder • SerializedCommit • RestartConvoy • FriendlyFire • DuelingUpgrades • FutileStall • StarvingWriter LL EL EE University of Wisconsin Madison

  22. System EE • Conflict Detection: Eager • Detect conflicts immediately • Version Management: Eager • New Value in-place, Restore on Abort - Abort + Commit • Conflict Resolution • Stall Requester • Abort Requester on possible deadlock (Conservative Deadlock Avoidance) compare with LogTM University of Wisconsin Madison

  23. T1 T2 load A load A store A STALL store A ABORT load A store A DuelingUpgrades • CAUSE • Conservative Deadlock Avoidance • EFFECT(S) • Unnecessary Stall • Unnecessary Abort • FIX • Acquire Store Permissions Early ? University of Wisconsin Madison

  24. T1 T2 T3 T4 load A store A load A load A load A load A StarvingWriter • CAUSE • Stall Requester • EFFECT(S) • Starvation • Load Imbalance • ≈ Reader/Writer Locks • FIX • Writer Priority ? University of Wisconsin Madison

  25. Pathologies EE • Pathology (% ExecutionTime) • Enhancements • DuelingUpgrades: Store-Set Predictor • StarvingWriter: Writer Aborts Readers immediately EE_HP EE University of Wisconsin Madison

  26. Enhanced EE ( ) EE_HP • Pathology (% ExecutionTime) • Performance EE_HP EE University of Wisconsin Madison

  27. Caveats / FutureWork LL_B • High Contention Scenarios • Better Conflict Resolution • Unidentified Pathologies • Better TM Workloads LL University of Wisconsin Madison

  28. Executive Summary • Performance Pathologies • Enhanced HTM systems Livelock University of Wisconsin Madison

  29. FutileStall DuelingUpgrades FriendlyFire RestartConvoy StarvingWriter StarvingElder SerializedCommit

More Related