climateprediction.net home page
Seemingly cp.net progress not saved resulting in BOINC thinking it is overcommitted

Seemingly cp.net progress not saved resulting in BOINC thinking it is overcommitted

Questions and Answers : Windows : Seemingly cp.net progress not saved resulting in BOINC thinking it is overcommitted
Message board moderation

To post messages, you must log in.

AuthorMessage
nsandersen

Send message
Joined: 12 Dec 04
Posts: 6
Credit: 518,551
RAC: 0
Message 19604 - Posted: 24 Jan 2006, 9:34:48 UTC

I like the BOINC idea although some of its quirks have given me problems earlier. Therefore I try to keep an eye on it now and then and report any oddities.

For a while the scheduling has behaved in what seems like an odd manner to me and I think it is due to the expected remaining time and/or the progress % for climatepredicition.net not being saved when the computer shuts down. This results in large fluctuations in the queue length in the following manner:

1) I turn on the computer in the morning and cp.net is at 0%, 750h expected before completion. BOINC decides the computer is overcommitted (this happens even if it has a week to finish only 4-5 hours of computation work in the form of smaller WUs in addition to the cp.net WU it is running, which has 8 months remaining to finish and is in no danger of not doing so).

2) BOINC finishes small WUs on an earliest-deadline-first basis until only the cp.net WU remains.

3) It then resumes the cp.net WU, which retrieves its position (currently about 70%), suddenly subtracting about 400h of CPU time from the queue. This results in it resuming round robin mode and downloading more small workunits to be ready to switch and continuing in this mode until I switch off the computer.

4) GOTO 1)


In other words, BOINC spends about half it\'s time or more in earliest-deadline-first mode, which is probably not quite fair to your project. Seemingly because the progress/estimated remaining time isn\'t retrieved upon startup, only when the cp.net WU is actually resumed.

I run BOINC 5.5.2 and hadsm3 v4.13 under Windows 2000.
ID: 19604 · Report as offensive     Reply Quote
Profile Keck_Komputers
Avatar

Send message
Joined: 5 Aug 04
Posts: 426
Credit: 2,426,069
RAC: 0
Message 19606 - Posted: 24 Jan 2006, 17:55:38 UTC

The latest recomended version of BOINC is 5.2.13, my personal recomendation is to run 5.2.8 or later.

I saw this problem, and more serious related ones, fairly frequently in the 4.xx clients. They have been reduced in the current client.
BOINC WIKI

BOINCing since 2002/12/8
ID: 19606 · Report as offensive     Reply Quote
nsandersen

Send message
Joined: 12 Dec 04
Posts: 6
Credit: 518,551
RAC: 0
Message 19774 - Posted: 30 Jan 2006, 12:04:09 UTC - in response to Message 19606.  

That fixes it, thank you. I didn\'t expect it to be BOINC related, as it only seemed to be CP.net that didn\'t recall it\'s progress upon startup.

ID: 19774 · Report as offensive     Reply Quote
nsandersen

Send message
Joined: 12 Dec 04
Posts: 6
Credit: 518,551
RAC: 0
Message 20659 - Posted: 23 Feb 2006, 11:42:55 UTC

And now it is back, unfortunately, with BOINC 5.2.13 and hadsm3 4.13..
ID: 20659 · Report as offensive     Reply Quote
Profile Thyme Lawn
Volunteer moderator

Send message
Joined: 5 Aug 04
Posts: 1283
Credit: 15,824,334
RAC: 0
Message 20721 - Posted: 24 Feb 2006, 13:51:37 UTC
Last modified: 24 Feb 2006, 14:27:40 UTC

I had a similar problem with one system the other day. I traced it to a problem with the result\'s deadline in client_state.xml

Somebody had set the clock back 10 years on the system and then back to what it should have been. The result\'s deadline was reset to over 100 years in the past, exactly as described in this thread. I never thought I\'d have to apply my own fix!

A bug report is heading towards the BOINC development mailing list.
"The ultimate test of a moral society is the kind of world that it leaves to its children." - Dietrich Bonhoeffer
ID: 20721 · Report as offensive     Reply Quote

Questions and Answers : Windows : Seemingly cp.net progress not saved resulting in BOINC thinking it is overcommitted

©2024 cpdn.org