SC is the International Conference for
 High Performnance Computing, Networking, Storage and Analysis

SCHEDULE: NOV 13-19, 2010

Characterizing the Impact of Soft Errors on Sparse Linear Solvers

SESSION: Research Poster Reception


TIME: 5:15PM - 7:00PM

AUTHOR(S):Sowmyalatha Srinivasmurthy, Manu Shantharam, Padma Raghavan, Mahmut Kandemir

ROOM:Main Lobby

The increase in on-chip transistor count facilitates achieving higher performance, but at the expense of higher susceptibility to soft errors. Data corruption in caches due to soft errors is a concern for long running scientific applications. We investigate the impact of soft errors for such applications that involve sparse iterative linear solvers such as the preconditioned conjugate gradient (PCG) method. We show that the reliability of PCG depends on how the soft errors are propagated though sparse matrix vector multiplication and we provide an in-depth analysis of the impact on convergence. We characterize the correlation between the numerical properties of the coefficient matrix and the performance degradation of PCG. We propose an energy-aware selective protection scheme. An initial evaluation of our proposed selective protection scheme indicates that our scheme, on average, improves the energy efficiency of caches by 15% and reduces performance degradation by over 60%.

Chair/Author Details:

Sowmyalatha Srinivasmurthy - Pennsylvania State University

Manu Shantharam - Pennsylvania State University

Padma Raghavan - Pennsylvania State University

Mahmut Kandemir - Pennsylvania State University

Add to iCal  Click here to download .ics calendar file

Add to Outlook  Click here to download .vcs calendar file

Add to Google Calendarss  Click here to add event to your Google Calendar

   Sponsors    IEEE    ACM