Hi there,
The title says it all. When I upgraded to version 6.6.20 I couldn't keep a supply or Rosetta WU. It won't download any new workunits until it is out of work units, and then it just downloads about 6 to 8 hours of work. I'm sharing my time between Rosetta (%60) and Seti (%40) It's set to keep a 2.7 day supply. I've reset the project, even detached and reattached, no change. Seti is working normally with about a 24 hours supply of work, Rosetta only keeps 1 or 2 workunits, at about 3 hours each.
I've upgraded to 6.6.28 with no change in the problem. Before the 6.6 series this never happened. Any ideas out there?

Forgot to mention this is on a dual core XP machine.

An update and more information.

It seems this is a problem relating to the combination of Seti and Rosetta, or maybe anytime multiple projects are running. If I stop downloading new work on Seti for a while, the Rosetta queue will fill up correctly. When I reenable Seti downloads it seems that Seti won't refill until it is out of WU, and then it just keeps one or two WU in queue even though I've set it for 4 days worth. So, it appears that there is always one project with a ton of WU, and the other has only one or two. Any ideas out there?

That could be a problem with the debt algorithm. If you want to reset the debt values to zero, there's a cc_config option now for doing that:

<zero_debts>
Set resource scheduling debts to zero. This lets you fix situations where client bugs have resulted in distorted debt values. Set it, run the client, then clear it (otherwise you'll start out with zero debts every time you run the client). New in 6.6.11

More information about cc_config.xml you can find here.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

Thanks for the suggestion, but I've already done that a few times. When that didn't work I detached Rosetta and then reattached. Still no improvement. It really seems like the current BOINC just lets whatever client is more aggressive fill up the queue instead of only allowing that client to fill it's portion of the queue.

The work cache which you have set is not per project, it will include all off the projects attached to your BOINC manager.
Is this your computer?
It doesnt show any workunits for rosetta completed within the last 30 days if it is.
Are you using any optimized applications for SETI? They have been known on my computers to mess up the debt system with other projects.

Is this your computer?

I don't think so, as that nArf is from Germany and not attached to SETI.
My guess is this narf, though the user name is Dale Kirstein at Rosetta.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

Yup, my workunits are listed under Dale Kirstein.

I guess what my concern is is that Boinc will let a project almost run out of WU, then download the full 3 days worth of work. This is even though I have it set for continuous connection. Right now I am down to 6 hours of work in Rosetta, and even when I force a connect it won't ask for new WU. When I finally run out of WU it will start downloading again. Of course, if there is a problem with the project I will easily run out of WU.

It is told that some newer BOINC versions had scheduling problems. By the way, 6.6.28 isn't the "current" one any longer, though the newest (6.6.31) is said to have problems with stopping the client. I can't tell from experience, since I still run BOINC v5 :-) So, since you aren't depending on CUDA, you could always retract to a version that handles the queues comfortly for you.

Gruß,
Gundolf

---

Computer sind nicht alles im Leben. (Kleiner Scherz)

If you are not cuda reliant then I would role back to 6.4.7 or 6.4.5. I am running Rosetta on all of my computers currently using 6.6.28 and the two versions above. I don't seem to have issues with any versions requesting work. However, I have noticed that sometimes because of bugs with the rosetta application and not the BOINC client, the Time Duration Correction factor for my PCs changes because of long running scans and therefore does not request new work for a while. It never runs out though.

It's against my nature to run anything but the latest version, but I'm thinking I'll need to roll back to an earlier version. I've tried 6.6.31, and it has the same behavior. Of course, I won;t know if this is ever fixed unless I keep trying the new versions.

If you go to the project tab and select the project and hit the properties button... what rDCF does it show? Regrettably the networking and computing stats are not included in there such as <on_frac> to get a more complete picture.

Sorry, I got tired if running out of work. I rolled back to 6.4.7 and now the work queues are back to normal. I'll just stay with this version I guess.