Overcommitted
4 Jun 2005 21:06:23 UTC
Topic 189289
(moderation:
I am running E@H with Boinc 4.43 for Linux and get two
workunits although my computer is overcommitted as
mentioned in the Boinc Manager Messages.
My computer can only handle one WU a week!
Overcommitted
)
I am running E@H with Boinc 4.43 for Linux and get two
workunits although my computer is overcommitted as
mentioned in the Boinc Manager Messages.
My computer can only handle one WU a week!
There is a known bug with 4.43. When a WU is started or restarted from disk, for the first few seconds the client eroniously believes that it has 0 CPU time remaining, and that it is therefore imperative to download more work from the project. I have checked in a fix which is incorporated into 4.44 that fixes this particular problem. I hope that the general symptom is not also caused by some other portion of the code.
BOINC WIKI
RE: I am running E@H with
)
Glad to know it is a bug, but it seems to happen only with E@H. It does not with seti@home.
Fellow Boincers: I have a
)
Fellow Boincers: I have a machine that apparantly wants to over commit,
machine# 210302. It is a P4 3.6HGz/945G platform running boinc 4.45. It is one of three indentically configured machines, the others operate as expected.
It has downloaded 29 work units due by 7/26 not including the two it is running now. It averages 2 completed units every 10 hours. While it can complete the work it has downloaded, I have noticed this machine will continue to download upto 16 work units per day if work is allowed. The problem here is that is way more work than this UP/HT machine can crunch.
Preferences: connect interval=1 day; E@H 100%, installed as single user, screen saver disabled, max CPU=8 (maybe could this be a factor)
I have selected "no new work" and will let it run dry then delete and reinstall the application (maybe try a later version) to see if this fixes the issue.
I know this was an issue in earlier versions but I haven't seen any issues with 4.45. If someone has a better suggestion, I'm open to it.
>Preferences: connect
)
>Preferences: connect interval=1 day; E@H 100%, installed as single user, screen saver disabled, max CPU=8 (maybe could this be a factor)
That means, this computer requests 8 Units in one time cause it expects that the computer can crunch 8 WUs at the same time!!!
HTH
C167
RE: Preferences: connect
)
C167, normally I would agree with you but that setting isn't generating similar results in the other machines. I believe max cpu setting is for MP machines with more than two processors. I have several DP/HT machines that look like they have 4 processors to E@H. Before I changed that setting, they would only work on two units at a time, now they operate 4 at a time. Regardless, there could still be a bug at startup related to that setting. I may try it lower and see if there is an effect.
The max CPU settings does not
)
The max CPU settings does not affect how much work is requested unless you actually have the CPUs.
Some versions (and 4.43 may have been one of them) had a bug where the work on hand was not counted correctly. In these cases, more work would be requested even though there was enough on hand.
BOINC WIKI
RE: I am running E@H with
)
I too have just attached to the project downloaded a work unit. The deadline for completion is seven days hence but the estimated processing time is over 229 hours, or over 32 hours per day. Some mistake, surely ?
RE: RE: I am running E@H
)
The download rules as set up are that each host will get one WU from a project even if it does not look as if it will complete on time. It is possible that your WU will complete faster and be returned on time. If it is not, I would suggest that you detach from this project as it would appear that your computer is too slow.
BOINC WIKI
Hi, Over the last month or
)
Hi,
Over the last month or two I have noted similar over commitment behaviour of the E@H project. I have three machine, Intel/AMD and WinXP/W2K combinations and all three machines exhibit this behaviour.
On my AMD64 3000 machine I have noted it on several occasions. Boinc downloads about 45hrs of work to be completed in 6 - 7 days with the resource share at about 8%. 8% of a week is about 14hrs! What happens then is that the scheduler eventually goes into panic mode (earliest deadline first, no work fetch allowed) and only processes E@H for hours on end to meat the deadlines.
Then, due to the huge negative debt (-150 000s) built up by E@H no WU is downloaded for days, if not weeks.
This problem occurs on all three machines and only with E@H, using v4.45, 4.70 and 4.71. Could it be that the E@H server calculates the number of WU's (ie processing time) incorrectly when Boinc requests a certain amount of work?
Thanks
V