JSU's Al'ona Furmanchuk Tops MCSR's List of 5 Most Efficient Parallel Researchers for 2^nd QTR FY 2007

In this issue of the Parallel-O-Gram we are happy to continue the "Top 5 Most Efficient Parallel MCSR Researchers" chart. This chart recognizes the most efficient parallel researchers of the quarter. Each quarter, we query the PBS Historical Database to determine which users have made the most efficient use of our parallel resources over that time period. By efficient use, we mean: which users are doing the best job of consistently running jobs which use all or most (and waste the least) of the CPU's they request for those PBS jobs. For our purposes, "wasted CPU cycles" refer to time during which a CPU is available to the job, but not actually being kept busy by the job. Theoretically, the most efficient job, or the job with the best utilization of the CPU's it has asked for, will consume 100% of the CPU time available to it. In actuality, several factors may limit that CPU efficiency. These factors include: large input/output activity (since the CPU is idle during disk access times); communication time between processes (especially on distributed memory architectures such as mimosa); and poorly parallelized algorithms. One of your responsibilities as a research user of MCSR is to try to ask for only those resources that you need, and to use what you ask for. One of our responsibilities, as your support staff, is to help discover both efficient and inefficient jobs, and alert you if your jobs are inefficient, but also to recognize when they are very efficient, in order to encourage more of the same.

So, without further delay, here are most efficient parallel researchers of the first quarter of FY 2007 (October - December 2006). Each of these researchers achieved over 90% utilization/efficiency over all parallel (>=2 processor, >=100 hours) jobs submitted on redwood and sweetgum. (Single-CPU jobs, which are the most efficient, are not considered.)

2^nd QTR FY 2007

Most Efficient Researchers, By Month, 2nd QTR

We encourage all researchers to regularly access the My MCSR portion of our web page, where you can view the efficiency of your currently running, or recently completed, parallel jobs.