No work assigments...

[Lazvon]

Is all volunteers to interact with the community… and is research grant money much more than volunteering really? Well, except for the lead researcher/professor (assuming they strike something they can monetize).

[SandyG]

Better today, so far. Fingers crossed as they say...

Will at some point start a new thread as to the state of affairs with FAH, always want to learn more, and understand how it works.

Sandy

[SandyG]

Seeing these starting to crop up again...

Code: Select all

 
04:32:14:ERROR:WU01:FS04:Exception: Server did not assign work unit
04:32:14:WU01:FS04:Connecting to assign1.foldingathome.org:80
04:32:15:WARNING:WU01:FS04:Failed to get assignment from 'assign1.foldingathome.org:80': No WUs available for this configuration
04:32:15:WU01:FS04:Connecting to assign2.foldingathome.org:80
04:32:15:WARNING:WU01:FS04:Failed to get assignment from 'assign2.foldingathome.org:80': No WUs available for this configuration
04:32:15:WU01:FS04:Connecting to assign3.foldingathome.org:80
04:32:16:WARNING:WU01:FS04:Failed to get assignment from 'assign3.foldingathome.org:80': No WUs available for this configuration
04:32:16:WU01:FS04:Connecting to assign4.foldingathome.org:80
04:32:17:WARNING:WU01:FS04:Failed to get assignment from 'assign4.foldingathome.org:80': No WUs available for this configuration
04:32:17:ERROR:WU01:FS04:Exception: Could not get an assignment

[Lazvon]

Yeah, had two in Ready State earlier this morning, ow a different one in Ready.

Guess we are near the end of work for F@H… and isn’t just some random problem to be fixed.

[Andre_Ti]

There is actually a lot more work to be done.
Now there are few large-scale projects. Servers cannot generate a large number of Work Unit Genes, because the number of clones most often does not exceed 500.
If they decide to assign smaller working units for powerful GPUs, then we will get a reduced PPD due to incomplete loading.

[Joe_H]

Also, please remember that in addition to being beginning of the academic semesters for most groups involved in F@h, today is a holiday in the US. Some researchers may be working anyways, but many are not.

[SandyG]

Also, please remember that in addition to being beginning of the academic semesters for most groups involved in F@h, today is a holiday in the US. Some researchers may be working anyways, but many are not.

Been going on for about a week now, so not sure that I would agree with the issue is a 'today' issue.

Bottom line is it really feels like a project that has a team that is absent, but is responsible for a HUGE amount of resources that people contribute.

I have been looking at stats on Extreme Overclocking and see a good amount of decline in generated points.

That being said, what can we do to help get things back on track?

Sandy

[[Ars] For Caitlin]

I had a very small (few minute) delay getting work right now as all 4 assignment servers claimed there was no work for my configuration (big Navi). Older NVIDIA box has no problems getting work.

[BobWilliams757]

Yeah, had two in Ready State earlier this morning, ow a different one in Ready.

Guess we are near the end of work for F@H… and isn’t just some random problem to be fixed.

From my understanding of the system used to assign, it's simply a matter of there not currently being enough work that suits the very fast GPU's with a lot of cores. It's the proverbial catch 22 in assignments. If they assign projects that will return poor PPD to big, fast GPU's people will be upset. If there are no projects for the same GPU's people will be upset. There will always be more work, and there is work for the less powerful GPU's now. But those fastest cards eat through projects so quickly it's easy to see how they get depleted.

There are close to 600K public CPU jobs showing in the servers, but many people don't fold on CPU so they will remain as they usually do.


I get it, and it's a bummer for those with the fastest gear. But I suspect as more and more people buy the faster gear it might get worse as well. I don't think they can just create the projects that suit the gear all the time.

[SandyG]

Yeah, had two in Ready State earlier this morning, ow a different one in Ready.

Guess we are near the end of work for F@H… and isn’t just some random problem to be fixed.

From my understanding of the system used to assign, it's simply a matter of there not currently being enough work that suits the very fast GPU's with a lot of cores. It's the proverbial catch 22 in assignments. If they assign projects that will return poor PPD to big, fast GPU's people will be upset. If there are no projects for the same GPU's people will be upset. There will always be more work, and there is work for the less powerful GPU's now. But those fastest cards eat through projects so quickly it's easy to see how they get depleted.

There are close to 600K public CPU jobs showing in the servers, but many people don't fold on CPU so they will remain as they usually do.


I get it, and it's a bummer for those with the fastest gear. But I suspect as more and more people buy the faster gear it might get worse as well. I don't think they can just create the projects that suit the gear all the time.

As I have said, I suspect that the project is not being managed well and as such no one is a Champion for it on the other side of the coin (Students, Project owners, etc.)

The idea of faster GPU's sitting idle should be a huge problem as more and more folks are crunching WU's with them, and it's not going to let up. I shut down my cluster of 25 or so x86 machines as a single RTX3060 likely did more then them all. So if the goal here is to process WU's and no work units available, seems like a huge problem.

I have said this before, in all cases I should NEVER see an idle GPU waiting for work or my assumption is that we have drained the queues of work. Work is work, and right now I see one of my RTX4090's sitting idle for the last hour.

I don't use any of my hardware for anything else, just FAH. So it rubs me the wrong way when these issues persist and affect so many contributors. Also, again to say it, very much a lack of transparency on the state of the union here at FAH (This is my opinion).

At some point I'll move on, but hoping that the issues are solved and things can become better, and that includes updated and related communications from those that are in the know to everyone else. I just don't see much of that. This is a great cause, but will be ruined by lack of communications and management.

Again all my opinion but not much being said by the folks running the show...

Sandy

[PaulTV]

Silly suggestion maybe (probably), it might well not be supported by the OpenMM library in the first place, but if there are only smaller jobs utilizing like half the GPU, why not run two (or more) simultaniously? I know, it's not working that way now. But considering GPUs seem to grow faster in number of cores than GPU jobs grow in atoms, it may be something really worth looking into, in a blue-sky thinking way of looking at it.

[dan39johnson]

Same here, a 3090 sitting idle for 4 hours now. I too bought this unit to do folding.

Last year something similar happened over a holiday weekend where they just ran out, but I too see the servers show they are full and serving up 0x22 jobs, so what gives?

My understanding was that a job could be configured to run on either cpu or gpu hardware.

Would also like to know what is going on with fah.

Dan

[dan39johnson]

And just like that, I got one! gremlins.

[[Ars] For Caitlin]

I've got BOINC on my boxes and it will nicely cede the card to F@H when there is work for Folding. I run Einstein (an astrophysics project that sifts through radio telescope data looking for various things) as the work units are not time dependent. Just one way to ensure my boxes are always doing some kind of science.

[smeghead!]

I've had ongoing issues with no work units for a while, assignment servers vary. I can leave the system trying to get a new WU for a log time to no avail. What does work every time tho is to reboot the computer.

At a guess I'd say the assignment server connection is stalling out & then not clearing the previous attempt from its cache so never proceeds to get a response; note that this is intermittent but once stalled a reboot is all that seems to fix it.

Does the V8 beta behave differently in this regard & would be worth trying?