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

SCHEDULE: NOV 13-19, 2010

Can the PGAS Programming Model Make Massive Analytics Possible?

SESSION: Research Poster Reception

EVENT TYPE: Poster

TIME: 5:15PM - 7:00PM

AUTHOR(S):Costas Bekas, Irina Fedulova, Alessandro Curioni

ROOM:Main Lobby

ABSTRACT:
Uncertainty quantification has become a crucial application in Science, Engineering and Business. In this context, computing the diagonal of inverse covariance matrices is of paramount importance. Standard techniques for this problem incur an exascale cost. In previous work (Bekas et al, WHPCF SC09) we have presented an algorithm that reduces this complexity by at least two orders of magnitude. We based our implementation on MPI and achieved 730 TFLOPS on 72 Blue Gene/P Racks. Here, we move this work forward and demonstrate that we can achieve competing levels of performance using a pure UPC based implementation. We show that the merits of the PGAS programming model lead to an elegant, simple and easy to follow implementation thus allowing for a huge increase in overall development productivity. To support these goals we have developed a new communication tracing library, DCMFTrace, which allows for seamless collection of tracing information of UPC programs.

Chair/Author Details:

Costas Bekas - IBM Research

Irina Fedulova - IBM

Alessandro Curioni - IBM Research

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