1 / 16

Exploring Transactional Event Profiling for Performance Improvement

This study presents the concept of Transactional Event Profiling in a Best Effort HTM environment, focusing on analyzing program execution to identify opportunities for performance enhancement through real-time adaptation. The collaboration between researchers from the University of Alberta and IBM Toronto and IBM TJ Watson led to the development of X-Ray Specs Event Profiling Analyzers, which enable detailed insights into transactional activities and program runtime behavior. By examining transaction rates, execution details, probe effects, and various modes of operation, the study highlights the importance of dynamic information for making informed decisions to optimize performance on architectures like the BG/Q. The future work involves enhancing the tool's capabilities to navigate log files effectively, support longer programs, and leverage event logging for transactional execution insights.

brad
Télécharger la présentation

Exploring Transactional Event Profiling for Performance Improvement

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. Transactional Event Profiling in a Best Effort HTM Matthew Gaudet Supervisor: José Nelson Amaral (University of Alberta) Collaborators: Amy Wang (IBM Toronto), Peng Wu (IBM TJ Watson)

  2. The Story.

  3. We need a pair of: X-Ray Specs

  4. Event Profiling Analyzers Program Log File Runtime Instrumented Hardware

  5. Analyzers Limited only by the power of your Imagination!

  6. Peeking into execution: 50000 cycles = 0.00003125 seconds

  7. Seeing Rates As they change

  8. Which transactions are active?

  9. Real TX Length Distributions

  10. Details matter LR Mode SR Mode

  11. Probe effects

  12. Using this Want to improve performance on BG/Q (Judicious) Serialization When? Too little? Bad performance Too Much? Bad performance Solution: Runtime Adaptation

  13. Understanding What’s going on?

  14. Solution: Correct decisions requires dynamic information

  15. Future Work • The ability to zoom and explore log files, like Visualizing Transactions • Support for longer, larger programs: In memory compression, partial log-dumps.

  16. Conclusion • Event Logging provides a useful view of transactional execution

More Related