World Community Grid - View Thread - FAH2 runtime estimate

There's a big difference in the runtime required for the current experiments on my 8-core i7-6700K Windows 10 system, with experiments 001770-002220 taking 5 times longer (mode ~4 hours 10 minutes) than 001319-001769 (mode ~50 minutes). This difference corresponds with the following statement on https://fightaidsathome2.cst.temple.edu/status.html:

Experiments 001770-002220 are the second set of simulations for the large-scale benchmark test of the asynchronous replica exchange protocols. Each WU simulation has MD steps of 50 thousands instead of 10 thousands for the Experiments 001319-001769.

When I fetch work I always seem to be sent tasks for the same experiment type, usually a number of times in succession. This is causing BOINC's runtime estimate for FAH2 tasks to fluctuate wildly because the experiment types both have the same <rsc_fpops_est> value (35601960763096.000000).

By way of example, after a sequence of 56 of the shorter experiment type I received 8 tasks for the longer type, with BOINC estimating that they'd only take 56 minutes. Those tasks were followed by 2 sets of 8 shorter ones, with BOINC estimating the runtime for the second set as 1:48:28.