Erasing/Reducing long term debt problems

Message boards : Number crunching : Erasing/Reducing long term debt problems

To post messages, you must log in.

AuthorMessage
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5691
Credit: 5,859,226
RAC: 0
Message 67532 - Posted: 1 Sep 2010, 19:37:27 UTC
Last modified: 1 Sep 2010, 19:38:02 UTC

I am running 4 projects on a quad core system right now.
My percentages are:

Rosie 36.11
Ralph 27.78
Milkyway 19.44
Einstein 16.67

Problem is I ran up some serious long term debt awhile back. I went into the Boinc projects folder and edited away all the long term and short term debt in the client state xml file as recommended by someone. This has not reduced the influx of Einstein tasks. I am getting about a 30:1 ration of Einstein vs Rosie in files. Einstein tasks run about 5:45 typically if its the gravity wave tasks. Radio telescope tasks take just over and hour to run but are not that frequent. Milkyway generates tasks that are 10hrs in run time.

So the question is, where else could debt numbers be stored if not in the client state file? How can I get Einstein to back off and get Rosie to come back with more tasks?
ID: 67532 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Murasaki
Avatar

Send message
Joined: 20 Apr 06
Posts: 303
Credit: 511,418
RAC: 0
Message 67533 - Posted: 1 Sep 2010, 19:54:53 UTC

Your BOINC manager may be trying to compensate for the current shortage of tasks at Rosetta by building up a long term debt on Einstein. Once Rosetta is back to normal capacity you may start to see a stabilising or even reversal of the current trend.

I am not an expert on BOINC's handling of debt, but it is notoriously bad at adjusting to micro-management. It may take the system some time to realise that Einstein is incurring debt now and switch around instead.

Have you noticed if your manager is calling Rosetta and not getting new work, or is it just not attempting to make contact at all?
ID: 67533 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5691
Credit: 5,859,226
RAC: 0
Message 67534 - Posted: 1 Sep 2010, 21:05:24 UTC - in response to Message 67533.  

Your BOINC manager may be trying to compensate for the current shortage of tasks at Rosetta by building up a long term debt on Einstein. Once Rosetta is back to normal capacity you may start to see a stabilising or even reversal of the current trend.

I am not an expert on BOINC's handling of debt, but it is notoriously bad at adjusting to micro-management. It may take the system some time to realise that Einstein is incurring debt now and switch around instead.

Have you noticed if your manager is calling Rosetta and not getting new work, or is it just not attempting to make contact at all?



Oh I had kind of forgotten Rosie was having capacity issues.
I have never seen that kind of problem since I came to this project.
Your idea is entirely plausible. I see the note on the homepage about workgen issues, so yeah, I guess it's grinding Einstein for awhile.
The last time Boinc contacted Rosie was 2 days ago and got 1 task.
Since then there has been no contact and that task has not come up in the queue yet. But again the workgen thing could be the reason.

Will wait until things are back to normal to see whats going on.
ID: 67534 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 4018
Credit: 0
RAC: 0
Message 67537 - Posted: 1 Sep 2010, 22:37:15 UTC

Right, you were probably crunching Einstein, not because of debt, but because of no available work from R@h. And now, perhaps you have such a supply of work, you will have to crunch it down before it goes out looking to download more.

The objective of the debt system is that over time, in spite of the shortage of work from R@h, you should see a "pay back" of Einstein debt to R@h. In other words, in a day or 2 you will probably see an influx of R@h work, and no new tasks requested from E@h.
Rosetta Moderator: Mod.Sense
ID: 67537 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5691
Credit: 5,859,226
RAC: 0
Message 67539 - Posted: 1 Sep 2010, 23:43:28 UTC - in response to Message 67537.  

Right, you were probably crunching Einstein, not because of debt, but because of no available work from R@h. And now, perhaps you have such a supply of work, you will have to crunch it down before it goes out looking to download more.

The objective of the debt system is that over time, in spite of the shortage of work from R@h, you should see a "pay back" of Einstein debt to R@h. In other words, in a day or 2 you will probably see an influx of R@h work, and no new tasks requested from E@h.



R@H has not looked for work in 2 days.
Einstein keeps stuffing up my work load.
Will wait until the problems with Rosie are solved and then see if things even out.
ID: 67539 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Mod.Sense
Volunteer moderator

Send message
Joined: 22 Aug 06
Posts: 4018
Credit: 0
RAC: 0
Message 67543 - Posted: 2 Sep 2010, 4:55:21 UTC

The BOINC client, especially with specific versions is notorious for over or under filling the work buffer. Once the buffer is too full, it doesn't request more work from any projects. You can check the date on the E@h website and see the last time your host contacted the project. Unfortunately it doesn't tell you the last request for work. A contact from the host may just have been to report completed work, perhaps it didn't ask for more. You'd have to review the messages on your machine to see that.
Rosetta Moderator: Mod.Sense
ID: 67543 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5691
Credit: 5,859,226
RAC: 0
Message 67544 - Posted: 2 Sep 2010, 9:19:17 UTC - in response to Message 67543.  

The BOINC client, especially with specific versions is notorious for over or under filling the work buffer. Once the buffer is too full, it doesn't request more work from any projects. You can check the date on the E@h website and see the last time your host contacted the project. Unfortunately it doesn't tell you the last request for work. A contact from the host may just have been to report completed work, perhaps it didn't ask for more. You'd have to review the messages on your machine to see that.



E@H is always in contact with its host.
R@H I forced contact and not reporting and not requesting new work.
1 task in queue, not yet running.
Will just wait and see if things will even out eventually in the next 2 weeks or so.
ID: 67544 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Sid Celery

Send message
Joined: 11 Feb 08
Posts: 2183
Credit: 41,726,991
RAC: 6,784
Message 67557 - Posted: 3 Sep 2010, 1:19:36 UTC

I had high hopes for Boinc after 6.10.56 and while it's certainly better than previous versions it hasn't quite solved it yet. If it isn't obvious, this is only an issue if you're joined to more than one project. I haven't seen any problem on machines only linked to Rosetta.

Greg: you could always set Einstein to "No new tasks" for a while until Boinc gets the hint and fills your buffer with Rosetta tasks, then revert Einstein back to normal and see how it plays out. A nudge in the right direction may help - or aimless tinkering may make it more confused. I don't know for sure, but worth a try at least.
ID: 67557 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5691
Credit: 5,859,226
RAC: 0
Message 67562 - Posted: 3 Sep 2010, 7:34:59 UTC - in response to Message 67557.  

I had high hopes for Boinc after 6.10.56 and while it's certainly better than previous versions it hasn't quite solved it yet. If it isn't obvious, this is only an issue if you're joined to more than one project. I haven't seen any problem on machines only linked to Rosetta.

Greg: you could always set Einstein to "No new tasks" for a while until Boinc gets the hint and fills your buffer with Rosetta tasks, then revert Einstein back to normal and see how it plays out. A nudge in the right direction may help - or aimless tinkering may make it more confused. I don't know for sure, but worth a try at least.



Sid, I was thinking the same thing. For the moment the buffer is full and nothing new is downloading. I reduced the buffer from 7 days to 5 days to reduce the number of tasks loading in, so it has plenty of work to chew through before there is room for new work. I will keep an eye on it and if Einstein loads in then it will be no new tasks for them for awhile.
ID: 67562 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote
Profile Greg_BE
Avatar

Send message
Joined: 30 May 06
Posts: 5691
Credit: 5,859,226
RAC: 0
Message 67592 - Posted: 5 Sep 2010, 0:11:55 UTC

aborted a bunch of einstein and let it catch up on some of that and some milky way. now rosie has 3 tasks running on my machine finally.

so it looks like things will slowly get back to normal after some more einstein tasks get out of the way.
ID: 67592 · Rating: 0 · rate: Rate + / Rate - Report as offensive    Reply Quote

Message boards : Number crunching : Erasing/Reducing long term debt problems



©2025 University of Washington
https://www.bakerlab.org