SC Conference - Activity Details

Improving the Performance of Parallel Scientific Applications Using Cache Injection

Edgar A. Leon  (University of New Mexico)
Doctoral Research Showcase Session
Thursday,  11:15AM - 11:30AM
Room 17A/17B
The memory wall adversely affects the performance of memory bound applications, particularly parallel scientific computations. Cache injection addresses this disparity by placing data into a processor's cache directly from the I/O bus. The effectiveness of cache injection on application performance is dependent on several factors including timely usage of data, the amount of data, and the application's data usage patterns. To improve application performance, I present policies to place incoming network data into the appropriate level of the memory hierarchy (L2, L3 or main memory). To this end, I demonstrate how cache injection compares to prefetching and how it may degrade application performance without an appropriate policy. Preliminary results from a set of policies tailored for MPI, show an overall improvement of 5% in the execution time of an actual application (AMG from the ASC Sequoia suite), and up to 15% improvement on specific stages of this code.
   IEEE Computer Society  /  ACM     2 0   Y E A R S   -   U N L E A S H I N G   T H E   P O W E R   O F   H P C