SPEAKER: Dean Tullsen
UC San Diego (CSE)
Simultaneous Multithreading machines fetch and execute instructions from multiple instruction streams to increase system utilization and speed up the execution of jobs. When there are more jobs in the system than there is hardware to support simultaneous execution, the operating system scheduler must choose the set of jobs to coschedule. This talk will demonstrate that performance on a hardware multithreaded processor is sensitive to the set of jobs that are coscheduled by the operating system jobscheduler. A mechanism will be presented that allows the scheduler to significantly raise the performance benefits of SMT architectures, just by careful selection of co-scheduled jobs. This is done without any advance knowledge of an application's characteristics, using sampling to identify jobs which run well together.
Other recent results from the Processor Architecture and Compilation Lab at UCSD will be also be discussed.