Obsolete work units
2 Mar 2005 8:45:26 UTC
Topic 188192
(moderation:
My computer is wasting time working on an outdated result (deadline is past), meanwhile probably neglecting the more important current results.
Is there any way to quit working on an outdated result?
Obsolete work units
)
With newer (experimental) clients 4.2x you can abort individual Results. With 4.19 clients there is no other way than to reset the project.
The somewhat hard way may work, at least it shouldn't go worse than resetting the project: stop the client and delete the slots directory (in the BOINC directory), then start it again. Unlike resetting the project this shouldn't cause another 11MB download.
BM
BM
> The somewhat hard way may
)
> The somewhat hard way may work, at least it shouldn't go worse than resetting
> the project: stop the client and delete the slots directory (in the BOINC
> directory), then start it again. Unlike resetting the project this shouldn't
> cause another 11MB download.
>
If you're running more than one project for BOINC, don't delete the whole Slots directory, or you'll lose all other projects as well. Just find the correct numbered slot, the one that has the Einstein executable in it. That's the slot you want to delete.
You also need to set your
)
You also need to set your connect to, to a lower number maybe .05 will be best for your system.......
Link to Unofficial Wiki for BOINC, by Paul and Friends
Somewhat related to "obsolete
)
Somewhat related to "obsolete WUs"
I think sometimes Einstein is overly optimistic when it fetches new WUs, especially if no other project is attatched.
Example 1:
http://einsteinathome.org/host/56367
Machine is AXP 2200+, running linux and needs 43.000s per WU. Cache is set to 0.5 days for school. Now after the first 2 WUs were done it had fetched new ones twice and accumulated a buffer of 15 WUs! That will take it 7.4 days or more, depending on regular load.
Example 2:
http://einsteinathome.org/host/15344
AXP2400+, win2k and 23000s/WU. Cache is 1 day for work. After it fetched new WUs twice during the last 2 days it build up a cache of 10 WUs = 2.66 days of computation time. Not es bad as the other one, but this machine is not working more than 10 hours a day, therefore the relatively small cache of 1 day.
... it would be so much easier for einstein to set a local "manual cache override" with a fixed (small!!) size for the cache.
Regards, [ETA]MrSpadge
Scanning for our furry friends since Jan 2002
well indeed E@H sends you new
)
well indeed E@H sends you new WUs based on estimates (how much time your PC is crunching) but if your PC suddenly fails to keep the rate it was crunching you will definately run out of time.
I've noticed that time to completion is too optimistically estimated (19 hours whereas in fact i complete in 24-25hours)
I wonder if the problem is
)
I wonder if the problem is the following. Einstein@Home estimates the CPU time needed to finish SCIENCE part of the application, but doesn't take into account the time taken to run the screensaver. Perhaps what is going on here is that the graphics thread is consuming more CPU time than the science thread.
Could people who have been getting too many WU report what fraction of the CPU time is going into graphics and what fraction into the science thread? This can be determined by running the task manager to watch total CPU time, and watching the BOINC manager to see the reported science thread time for the same period.
Bruce
Director, Einstein@Home
I asked my mate to look into
)
I asked my mate to look into this. As far as I can tell the first machine uses the linux console client, so I don't think there will be screensaver or graphics active.
OK, the 2nd box with win2k doesn't use BOINC as a screensaver at all and doesn't have the "show graphics" active, just pure crunching.
Regards, MrS
Scanning for our furry friends since Jan 2002
Update: there's no graphics
)
Update: there's no graphics active on the linux box. But it turns out that it keeps using the prefs for home (cache 2 days) instead of the location being set to "school". We can't figure out yet, why this is the case. The seperate prefs for school exist for Einstein and BOINC in general.
Another thought: in einstein you set a minimum cache size, while in the predictor options you can set a cache of min to max. Currently our accounts have no max value set.
Regards, MrS
Scanning for our furry friends since Jan 2002
> Update: there's no graphics
)
> Update: there's no graphics active on the linux box. But it turns out that it
> keeps using the prefs for home (cache 2 days) instead of the location being
> set to "school". We can't figure out yet, why this is the case. The seperate
> prefs for school exist for Einstein and BOINC in general.
>
> Another thought: in einstein you set a minimum cache size, while in the
> predictor options you can set a cache of min to max. Currently our accounts
> have no max value set.
>
> Regards, MrS
>
Predictor has outdated web code.
BOINC WIKI
OK, this is getting a bit
)
OK, this is getting a bit off-topic, but it still leads to obsolete WUs.
I don't know whether this is a known and general problem. During the last day I got convinced that for Einstein the "work" and "school" prefs are completely ignored! "Home" is the default profile. I have Home with 2 days cache, work 1 day and school 0.5 days (+ other differences). Oberserved behavior:
- "work" machine ignores resource share for "work"
- changed my "home" machine to "work" -> no change in resource share
- "school" machines cache more like 2 days than 0.5 days (when not as extreme as my 1st example above)
- "work" machines cache more like 2 days than 1 day
- my mate sent me parts of the log of one of the "school" machines, it says that it's using the "home" prefs
I have seperate prefs for "home", "work" and "school" set up for "einstein" and "general". As a quick work-around I lowered the home cache to 1 day, but really don't want to go lower for the dial-up machines.
Regards, MrS
Scanning for our furry friends since Jan 2002