Look into the Future
The ORNL research indicates significant progress in heterogeneous distributed computing, focusing on Tera-scale systems. The potential for PC clusters to dominate high-performance computing (HPC) remains uncertain, as market trends do not favor the required node density. Despite challenges, Linux may emerge as a viable HPC OS, reflecting positive signs from ASCI results. However, concerns about reliability, maintenance, and scalability persist, with system failures and recovery issues highlighted. The outlook for achieving high efficiency from Tera-scale systems remains cautious as the research continues.
Look into the Future
E N D
Presentation Transcript
Ask me a Yes/No question. ORNL Heterogeneous Distributed Computing Research Look into the Future Reply Hazy Try Again
As system is updated year to year or you get Multiple clusters ORNL Heterogeneous Distributed Computing Research Will Tera-scale systems be Heterogeneous? Yes, Definitely
In the LOW END ORNL Heterogeneous Distributed Computing Research Will PC clusters Dominate HPC? You May Rely On It
Market isn’t driving PC node density To level needed by Tera-scale Cplant = ASCI Red ORNL Heterogeneous Distributed Computing Research Will PC clusters make the TOP “5”? Cannot Predict Now
Even though Linux is not an ideal choice for HPC Desktop is the driving market ORNL Heterogeneous Distributed Computing Research Will Linux emerge as The HPC OS? Signs Point to Yes
based on ASCI results so far May as well buy five 1TF systems than one 5TF system ORNL Heterogeneous Distributed Computing Research Can I get reasonable % of peak on my Tera-scale system? Outlook Not So Good
MTBF is Shrinking Rapidly w/ Large COTS clusters ORNL Heterogeneous Distributed Computing Research Will the system stay up long enough to complete my simulation Very Doubtful
Scalable Detection, Notification, and recovery are open research ORNL Heterogeneous Distributed Computing Research Will the system automatically recover my job? Don’t Count On It
You don’t need a Crystal ball to see the software needs for tera-scale systems ORNL Heterogeneous Distributed Computing Research Summary Application runs slower than expected system fails before it completes and doesn’t recover the app.