Richard Haselgrove wrote: > The script compares the credit awarded to the current cohort of machines > under that project's active credit-scoring scheme (flop counting, in the > case of SETI), with the credit that would be awarded to the *same* cohort of > machines under the benchmark*time 'cobblestone' scheme. Then adjusts > accordingly, with smoothing and median-taking to reduce the effect of > outliers.
If there is a problem with Eric's script, it comes from the definition of a cobblestone itself. Cobblestones are defined in terms of two standard, time tested benchmarks. We can argue forever on what the benchmarks actually mean, but that is a different discussion, because the value of a cobblestone is defined by Whetstones and Dhrystones. Volunteers complained about benchmark * time because three (SETI) results would come back with three different credit claims, the lowest and highest are discarded, and credit was granted. Sometimes it was high, sometimes low, but on average, it was right. Counting FLOPs is repeatable, but it isn't as accurate, because the application has a different mix of instructions than called for in the definition, and all FLOPs are not created equally. Eric's script takes a secondary standard (FLOPs) and relates it back to the definition. Smoothing is needed as you state, so that short term variations don't swing credit around excessively. Richard, as you suggested, a rising tide lifts all boats. The benchmark speed goes up, the duration goes down, and they more-or-less cancel. Ultimately, if we try to move away from Whetstones and Dhrystones completely, we are changing the definition of a cobblestone. -- Lynn _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
