Running, high priority
14 May 2009 17:22:17 UTC
Topic 194345
(moderation:
Lately I only receive Einstein WU's which have to be ready on very short notice. Therefore the Boinc manager starts those first with high priority. So far so good.
But, when they some of them are done, I receive new ones also with a short time to crunch. Result: I am only crunching for Einstein in high priority mode. Other projects don't get a chance, until their dates have almost arrived as well.
I also want to crunch for other projects. Now I have to suspend Einstein for a couple of hours manually so I can also have some other work done.
Am I the only one who has this, or is it a new strategy of Einstein@home to have more work done this way?
______
DeMus
Running, high priority
)
What are the resource shares for your different projects?
Check the debts the projects have piled up. Those are listed in the popup properties (I'm still running 5.10.45, so I can't tell you exactly how to get them).
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: What are the resource
)
Hello Gundolf,
with resource shares you mean the percentage of time the manager can use this or that project?
Well, I set Seti to 400, Einstein to 50 and Rosetta to 50, giving me
Seti 80%, Einstein and Rosetta each 10%.
My orignal project was (and still is) Seti. Around New Year I added Einstein and Rosetta for the times Seti has no work to do.
Now Einstein is completely taking over.
What do you mean with the debts?
______
DeMus
RE: What do you mean with
)
If you search for "debt" on the BOINC FAQ site, you get (among others):
You can find those values also in the client_state.xml file in your BOINC data directory. Search for the string "debt". There should be two values per project.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: RE: What do you mean
)
Danke für die Antwort.
For Rosetta I have:
0.000000
-125780.768618
For Seti:
0.000000
0.000000
For Einstein:
0.000000
-156647.328850
Because the values for the Seti project are both 0 it means it won't fetch new units and it won't crunch units? Is that what it means?
But for all 3 projects the short_term-deb is 0. Then why it wants to crunch only Einstein units? And why do I only get new units with a very short expiry date?
______
DeMus
RE: Because the values for
)
No, in recent BOINC versions, the debt mechanism was changed: the maximum value is zero and debts are shown as negative values. So, in theory, SETI should be crunched before all else.
I don't know either. Perhaps you should set all projects to No New Tasks, empty your cache (don't forget to report all uploaded tasks) and reset all projects. I have read elsewhere that the debt transition not always goes smoothly.
What do you mean with "short"? All my Einstein tasks (both S5R5 and ABPS) have a deadline of 14 days; that's plenty.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
You can reset the debts far
)
You can reset the debts far more easily than that. Use cc_onfig flag. Stop BOINC's connected client, exit the manager, restart it (or reboot) and the debts will be reset. Yoiu cannot just re-read the cc_config file to reset the debts.
remove the flag (or set it to 0...
Virtually all of the 6.6.20 have LTD and work fetch problems. The only question is when are you going to be hit with them. Not if ... and what it will look like...
The MOST COMMON "look" is that your queue is shorter than normal and has a poor mix of work.
On my i7 I usually have to reset debt about ever 2-3 days... less often on my Q9300.
Part of the problem *I* have there is the fact that I hae 2 GTX295 cards in the i7 so, that causes the debts to get out of whack very quickly.
I have not yet looked at the code closely... yet ... on my to-do list ...
Not that I can seem to get anyone to listen ...
RE: You can reset the debts
)
To clarify this: It would be the option.
To use it, you must create a text file named cc_config.xml in your BOINC data directory (if it doesn't exist already). The data directory is mentioned in the message tab of BOINC manager.
With a text editor, enter (copy&paste) the following into that file:
1
There are no lower case 'L' characters in that code, only digits '1'. More information to cc_config you can find here.
Then continue as Paul D. Buck directed.
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: RE: And why do I only
)
What I mean with short is the time you mention, so short that the manager decides to give it high priority all the time. It makes my computer do only work for Einstein and not for the other 2 projects, including my main project which is still Seti. Sorry for that last remark.
I will empty my cache and reset the project. The other way is something I don't like to do since it sounds "dangerous". Sorry Paul.
______
DeMus
RE: RE: RE: And why do
)
Well the problem with using a high bias resource share with later 5x CC and higher (the screwy fetch and scheduling for 6.4 and above aside) is the policy to keep the overall cache full at all times means that if the primary is not available for any reason you will in all likelihood draw from what should be an LTD 'ineligible' (meaning a relatively large negative value) one unless it is already full.
IOWs, it is highly unlikely that you would ever reach your desired resource share on SAH because they have too many unexpected short outages, and then of course there is the regular weekly one. That one you can 'schedule out' for the most part with preferences, but there isn't anything you can do about unscheduled ones (except micro manage debt as Paul suggested).
Be advised though that resetting projects has some undesired side effects as well, such as resetting all your Time/Performance metrics back to the default. This means BOINC has to go thorough 'recalibrating' them again and that causes issues in and of itself. One other catch with resetting. In order to get all the debts to zero, you have to reset all your projects. Not just EAH alone, because the CC will redistribute the remaining total debt over all the projects. IMO, if you're going to do that it's far easier to shut down BOINC and edit the debt values in client_state.
The reason that EAH (or maybe even Rosetta in some cases) seems to take over is that in progress work scheduling is also based on resource share. So in essence you have to mulitply the average runtime your host takes on a 'backup' project's tasks by the RS ratio (in your case 10), and that's the time figure the CC is using when it decides who has to run now to be able to make deadline.
I experimented extensively with this last year, and with SAH as a primary the best I could reach for effective Resource Share was about 75% (IIRC) with the RS split set to be 99.9% to SAH. I came to the conclusion that SAH makes for a much better backup project than a primary, if keeping lockstep with your set RS is what you hold most dear. ;-)
Alinator
RE: RE: RE: RE: And
)
Remember there's always BOINC Debt Viewer, which can reset debt as well as display it. You just have to shut down the BOINC client yourself.