University of Cambridge > Talks.cam > Computer Laboratory Computer Architecture Group Meeting > Using Processor Hardware Counters in Picking the Optimal Work-Stealing Policy

Using Processor Hardware Counters in Picking the Optimal Work-Stealing Policy

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Prof Simon Moore.

Randomised work-stealing is a distributed dynamic load balancing scheme that has been proven to be optimal with tight space bounds for a large number of recursive task-parallel programs. Under work-stealing when a parent thread spawns a child, worker will make parent thread available to idle workers to steal it. However, if parallelism manifested in program is iterative then the steals from idle workers will be serialised and therefore had a negative impact on scalability of the scheduler. In this case we would have been much better off if we chose to continue executing the parent thread and make a child thread available for stealing.

In this talk I will introduce tools I have developed to characterise different types of task-parallel benchmarks and present the insights how to exploit now ubiquitous hardware counters to dynamically pick and switch between the optimal policies during the program’s lifetime.

This talk is part of the Computer Laboratory Computer Architecture Group Meeting series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

© 2006-2024 Talks.cam, University of Cambridge. Contact Us | Help and Documentation | Privacy and Publicity