Questions and Answers : Windows : Time to completion always about 14 hours before workunit start
Message board moderation

To post messages, you must log in.

AuthorMessage
David Ball

Send message
Joined: 29 Sep 04
Posts: 6
Credit: 127,342
RAC: 0
Message 8382 - Posted: 10 Jul 2005, 19:00:43 UTC

I have 2 very different machines running LHC. One always says a little over 14 hours and one always says a little under 14 hours for time to completion(TTC) until they start running. Both machines are running BOINC 4.45 and have BOINC jobs set to remain in memory when preempted. They each have 2+ GB of swap space and both run 3+ BOINC projects.

Machine # 1
OS: Windows XP Pro
CPU: Celeron 2.3
Ram: 1 GB
starting TTC: 14 hours +
Actual TTC: about 4 hours
Internet Connection: 56K Dial On Demand

Machine # 2
OS: Windows Server 2003
CPUS: Dual Xeon 2.8 HT
Virtual CPUS: 4
Max BOINC jobs running at once: 2 (they seem to get distributed evenly across the 4 virtual CPUs according to the performance graphs for the 4 virtual CPUs. With 2 BOINC jobs running my total CPU usage runs 50% - 55%.)
Ram: 1 GB
starting TTC: just under 14 hours
Actual TTC: about 3.25 hours
Internet Connection: 100 Mbps Ethernet


Any idea why the TTC is off by +10 hours until the job starts. Does this keep boinc from doing as much work on LHC since these jobs say they will take 14 hours until they actually get a little run time. I presume that BOINC bases how much work it fetches at once for each project on the TTC of jobs in the work queue (among other things) so it fetches less work at a time and work might not be available when it finally figures out that the work unit is 10 hours shorter than expected. This can also cause BOINC to go into it's time critical scheduling method because it thinks it has more work in the queue than it really does.

-- David

ID: 8382 · Report as offensive     Reply Quote
Heffed

Send message
Joined: 2 Sep 04
Posts: 71
Credit: 8,657
RAC: 0
Message 8387 - Posted: 10 Jul 2005, 21:21:52 UTC

The newer versions of the CC currently in alpha testing are working to help out with this issue.

The estimated time to completion becomes closer to the actual processing time as you process WUs. If your actual completion time is lower than the estimate, the estimate gradually lowers. It lowers slowly, in case of fluke WUs like noisy WUs from S@H.

This feature will be somewhat less effective with LHC WUs, due to the variable sizes sent out. Your estimated time will most likely always hover around the completion times of the longest WUs, because the value raises rapidly if you get a longer running WU.
ID: 8387 · Report as offensive     Reply Quote

Questions and Answers : Windows : Time to completion always about 14 hours before workunit start


©2024 CERN