Benchmark Results

With our initial cluster of 8 dual-processor G5s, we optimized FSL to take advantage of multiple processors with the help of Sun Grid Engine. This allows for dramatic decreases in time taken for analyses.

The FSL analyses below were run both with cluster-optimized-FEAT as well as on a single dual-processor G5 with regular FEAT. The bars represent total time for a full analysis, both first-level and higher-level. The cluster generally decreases total analysis time by eighty to ninety percent, often a difference between running an analysis over lunch versus running it overnight.

The data shown here are primarily from the 8 XServe configuration (fourteen active cpus), prior to the introduction of a RAID array, and do not yet take advantage of alitvec. The graph at lower right is in the current configuration. Seventeen cpus were active in that analysis.

<< More on the cluster