climateprediction.net home page
Resource share problem

Resource share problem

Questions and Answers : Windows : Resource share problem
Message board moderation

To post messages, you must log in.

AuthorMessage
DanNeely

Send message
Joined: 26 Aug 06
Posts: 4
Credit: 1,567,994
RAC: 1,104
Message 24158 - Posted: 2 Sep 2006, 20:45:37 UTC

My machine is an A64x2. Several days ago I added a single cpdn work unit to my project list.

I have 3 other projects all with equal resource share. Of these LHC doesn\'t have any work, and simap is suspended on this PC (I\'m running it on an older box, and only added it to this one for stats display). My 4th project is einstien. It and cpdn are running at 100% on alternate cores.

The problem is that since I started the cpdn unit, boinc is refusing to download any work for einstien because my machine is over commited, and my cache has dropped from it\'s normal level of several days to 8hrs. Einstien itself is running normally, and has work available. Once my cache runs out, my second core will be idle. Short of aborting the cpdn unit and going back to 100% einstien, how can I get boinc to provide work for both of my cores.
ID: 24158 · Report as offensive     Reply Quote
Les Bayliss
Volunteer moderator

Send message
Joined: 5 Sep 04
Posts: 7629
Credit: 24,240,330
RAC: 0
Message 24160 - Posted: 2 Sep 2006, 22:28:04 UTC

Leave it alone for a while.
When BOINC works out the time-to-completion of the climate model, (which is a couple of thousand hours), it will know that it can handle it OK, and get more work from the other projects.
One of the problems now for climate models, is that BOINC 5.* is optimised for short WUs, and is hopeless with the climate models.

ID: 24160 · Report as offensive     Reply Quote
Profile Keck_Komputers
Avatar

Send message
Joined: 5 Aug 04
Posts: 426
Credit: 2,426,069
RAC: 0
Message 24162 - Posted: 2 Sep 2006, 23:01:34 UTC - in response to Message 24160.  

Leave it alone for a while.
When BOINC works out the time-to-completion of the climate model, (which is a couple of thousand hours), it will know that it can handle it OK, and get more work from the other projects.
One of the problems now for climate models, is that BOINC 5.* is optimised for short WUs, and is hopeless with the climate models.


Also the client should not let a CPU run dry (for long). When it does it will get another einstein task even if it still thinks it is overcommitted.
BOINC WIKI

BOINCing since 2002/12/8
ID: 24162 · Report as offensive     Reply Quote
DanNeely

Send message
Joined: 26 Aug 06
Posts: 4
Credit: 1,567,994
RAC: 1,104
Message 24164 - Posted: 3 Sep 2006, 0:28:10 UTC - in response to Message 24162.  

Leave it alone for a while.
When BOINC works out the time-to-completion of the climate model, (which is a couple of thousand hours), it will know that it can handle it OK, and get more work from the other projects.
One of the problems now for climate models, is that BOINC 5.* is optimised for short WUs, and is hopeless with the climate models.


Also the client should not let a CPU run dry (for long). When it does it will get another einstein task even if it still thinks it is overcommitted.


I\'ve had enough weekend isp outages that downloading a new work unit only when the old one runs out isn\'t an acceptable situation. If I can\'t keep a standby workload this isn\'t going to work.

ID: 24164 · Report as offensive     Reply Quote
DanNeely

Send message
Joined: 26 Aug 06
Posts: 4
Credit: 1,567,994
RAC: 1,104
Message 24165 - Posted: 3 Sep 2006, 0:30:28 UTC - in response to Message 24160.  

Leave it alone for a while.
When BOINC works out the time-to-completion of the climate model, (which is a couple of thousand hours), it will know that it can handle it OK, and get more work from the other projects.
One of the problems now for climate models, is that BOINC 5.* is optimised for short WUs, and is hopeless with the climate models.



I\'ve left it alone for about 5 days now, and the ETA doesn\'t appear that far off. Boinc\'s calling the WU 1700hrs, dividing the hours spent so far by the percentage done, it looks like about 2000. IF telling boinc it will actually take longer will do any good, I can edit the DCF in client_state.xml
ID: 24165 · Report as offensive     Reply Quote
Profile MikeMarsUK
Volunteer moderator
Avatar

Send message
Joined: 13 Jan 06
Posts: 1498
Credit: 15,613,038
RAC: 0
Message 24171 - Posted: 3 Sep 2006, 9:38:03 UTC - in response to Message 24164.  

...
I\'ve had enough weekend isp outages that downloading a new work unit only when the old one runs out isn\'t an acceptable situation. If I can\'t keep a standby workload this isn\'t going to work.



If you want a standby einstein work unit:

* Set \'no more work\' against the climate project
* Suspend the current climate work unit

Boinc should start downloading an einstein unit.
As soon as it starts, resume the climate work unit. Alternatively, if you want a bank of several einstein work units, keep suspending them when they start to download, until you have sufficient.
I'm a volunteer and my views are my own.
News and Announcements and FAQ
ID: 24171 · Report as offensive     Reply Quote
Profile Keck_Komputers
Avatar

Send message
Joined: 5 Aug 04
Posts: 426
Credit: 2,426,069
RAC: 0
Message 24173 - Posted: 3 Sep 2006, 9:56:57 UTC - in response to Message 24165.  

IF telling boinc it will actually take longer will do any good, I can edit the DCF in client_state.xml

You would want to reduce the DCF if you go this route. That way the client will think there is less danger of missing the deadline and may request more work. Make sure you stop BOINC completely before editing the file.
BOINC WIKI

BOINCing since 2002/12/8
ID: 24173 · Report as offensive     Reply Quote
DanNeely

Send message
Joined: 26 Aug 06
Posts: 4
Credit: 1,567,994
RAC: 1,104
Message 24175 - Posted: 3 Sep 2006, 15:37:18 UTC - in response to Message 24173.  

IF telling boinc it will actually take longer will do any good, I can edit the DCF in client_state.xml

You would want to reduce the DCF if you go this route. That way the client will think there is less danger of missing the deadline and may request more work. Make sure you stop BOINC completely before editing the file.


I\'ve tried setting my DCF to extremely low numbers, but the ETA stops scaling downward at ~7%, which is not short enough to get boinc to resume automatic download of einstien work.


I was able to force the DL of ~8 coredays of einstien work by doing the suspend/resume trick.
ID: 24175 · Report as offensive     Reply Quote

Questions and Answers : Windows : Resource share problem

©2024 cpdn.org