GW S6Bucket Follow-up #1 observations
2 Dec 2014 17:58:08 UTC
Topic 197842
(moderation:
To give a full name Gravitational Wave search S6Bucket Follow-up #1 second try work release began a few hours ago as announced by Bernd Machenschalk in the Technical News thread. This second release followed a brief initial release which was terminated after several reports of a high incidence of early terminations, tracked to generation of WUs with wrong parameters because of a bug.
All three of my primary hosts downloaded second batch follow-up units during the night, and I suspended other work to expedite initial view of five units.
I've started this thread hoping others will share observations.
My own first observation is that on all three of my hosts the initial BoincMgr-reported estimated time to completion estimates are very high compared to indicated progression through at least the first half of reported unit progress. The mismatch appears to be about an order of magnitude (i.e. 2 to 4 day initial completion estimates for work which seems on current progress likely to finish in about 7 to 11 hours). All three hosts have current DCF in the 1.1 to 1.7 range, and currently estimate v1.05 FGRP4 work completion times no more than 50% over the truth. So one way or another I suspect this work is being distributed with parameters which will lead to substantial initial completion time over-estimates on most hosts.
Of course reports from other users would help clarify this point.
If anyone here besides me uses Process Lasso or something similar to adjust task-specific CPU affinity and priority, they may need to respond to the current executable name for these tasks hsgamma_FGRP4_1.05_windows_intelx86__FGRP4-Beta.exe
As this task name contains neither the substring Einstein, nor SSE2, it might not be detected by some user templates.
GW S6Bucket Follow-up #1 observations
)
My first S6Bucket Follow-up #1 is 20% complete at 1 hour 37 minutes, or 8 hours 5 minutes total (estimated). The "Time Left" shows as 1 day 21 1/2 hours.
As for the FGRP4-Beta, they are completing in a little over 8 hours also, and the estimates are by now more or less correct.
This is running 4 cores on a Haswell i7-4771, Win7 64-bit.
One of my first S6Bucket
)
One of my first S6Bucket Follow-up tasks has reached 20.9% done in 1h47m, I estimate it to finish at about 8.5h if the progress is linear. Boinc's estimate is showing 65.5h left but falling with about a minute every 7 seconds. =)
Unstarted tasks in my cache show 68h40m estimated time with a DCF of 1.8558 and would be estimated at 37 hours if the DCF was 1. This points to a overestimation of about 4.35 compared to the task described above on my system.