Missing driver just, keeps trying to get tasks.

[Peter_Hucker]

The most I can see is the summary information on the Server Stats page - https://apps.foldingathome.org/serverstats. It lists WU totals for the various folding cores, A8 and A9 are CPU jobs, the FE core is a benchmarking test core mostly used to test new GPUs, and the rest are GPU WUs. I have lost track of which cores and projects they have found issues with using various series of AMD GPUs. One of the later RDNA series is blocked from another set of projects because they perform very poorly on them.

Presumably the info I seek is there, it's just a matter of who is allowed to see it. Are you not as high up as I thought?

[Joe_H]

I am the remaining volunteer admin for this forum. There are several moderators here as well, all volunteers. I am aware of discussions about various problems with some GPUs and projects through some reports here, and from participating from time to time in internal testing. Though it has been some time since I did any GPU testing, currently only running Macs and doing CPU folding.

Now the info you seek is probably there, but most of it is only accessible to the researchers and the developers of the client and the folding cores.

[Peter_Hucker]

Ok. Let me know if I can help testing (with my two old card types!) Got a good variety of CPUs though, 10 machines from 2 core to 24 core. Will be buying a new card soon, about £100, if you want me to get a particular one let me know.

[muziqaz]

The client can't be coded to detect and suggest fixes for every corner case. If you have a spare $100k to give them to hire a second full time developer, then that could happen. Or someone might look at the open source code for the client and suggest additional code.

Hawaii and Tahiti are from two different generations of GPU chips by AMD. They mostly have similar capabilities, but are not an exact match. That has led to issues at times. Not all AMD cards are enabled for the same projects.

Any way I or you can see how many tasks for each type of GPU in the queue so I can confirm it's just a lack of work? There was continuous work a year ago when I last used that machine on folding.

It is lack of work. End of discussion.
You don't come in to a project with decades old hardware and demand or expect that it is supported and has work for it.
We are approaching the date when I will remove 1st gen GCN GPU completely from supported list.

[Peter_Hucker]

Please don't write "end of discussion", it's quite rude.

I am not demanding you make work for it, I'm asking if that's the problem and why it's a problem considering it's no older than one I'm happily using.

And why do you assume everyone is rich and has brand new hardware? I'm not rich, and I like using things until they break, call it environmental friendliness. Electricity (can be) a renewable resource. Building circuit boards again and again, we use up resources, I bet we never truly 100% recycle all the precious metals in them.

[calxalot]

I think WUs keep getting bigger and more demanding. So older hardware may age out sooner than one might expect when they can’t meet deadlines. Deadlines are not arbitrary. Research projects need to be completed in a reasonable time frame.

[muziqaz]

Please don't write "end of discussion", it's quite rude.

I am not demanding you make work for it, I'm asking if that's the problem and why it's a problem considering it's no older than one I'm happily using.

And why do you assume everyone is rich and has brand new hardware? I'm not rich, and I like using things until they break, call it environmental friendliness. Electricity (can be) a renewable resource. Building circuit boards again and again, we use up resources, I bet we never truly 100% recycle all the precious metals in them.

Unfortunately projects like this are elitist by design. Current hardware dictates the advances, not the old hardware. In regards to moral dilemma of human race resorting to consumerism should not be discussed here, since this place will not solve that issue. This is distributed computing project which prefers fast and modern hardware. And this is not considered a problem at all. If you can contribute, that is fine, if you cannot that is also totally fine. What rubs people off is new users coming in and assuming that things which do not fit them are the problem on our end, which we don't know how to fix.
Some of us have been with the project from early 2000s. We know what our problems are and what they are not. hardware support isn't the problem. There is no need for newcomer outsider to come in and start thinking otherwise. If your old hardware does not get any work, or gets little work, that is by design and not because we personally are elitist and hate people who love to recycle.

[Peter_Hucker]

I think WUs keep getting bigger and more demanding. So older hardware may age out sooner than one might expect when they can’t meet deadlines. Deadlines are not arbitrary. Research projects need to be completed in a reasonable time frame.

I can accept if my hardware doesn't meet "timeout", I'd stop using it. The loss of points is presumably to encourage users to stop using very slow cards, or ones which aren't on very long at a time.

But my "slow" cards are easily finishing on time.

[Peter_Hucker]

Unfortunately projects like this are elitist by design. Current hardware dictates the advances, not the old hardware. In regards to moral dilemma of human race resorting to consumerism should not be discussed here, since this place will not solve that issue. This is distributed computing project which prefers fast and modern hardware. And this is not considered a problem at all. If you can contribute, that is fine, if you cannot that is also totally fine. What rubs people off is new users coming in and assuming that things which do not fit them are the problem on our end, which we don't know how to fix.
Some of us have been with the project from early 2000s. We know what our problems are and what they are not. hardware support isn't the problem. There is no need for newcomer outsider to come in and start thinking otherwise. If your old hardware does not get any work, or gets little work, that is by design and not because we personally are elitist and hate people who love to recycle.

Fair enough, if the system said to me "sorry, Hawaii cards don't get work any more because it's too much trouble for the programmers to write code for every single card". But I didn't know that was the problem, I saw my Tahitis getting work and since they're the same age, assumed it was an error. Perhaps "no work available for Hawaii just now, please wait to see if there is later", then I'd know it was a work queue problem, not an error.

Funnily enough, it just got one. I guess some professors are running a different code set.

Now I just need a way for Boinc to notice folding is running and pause. Is there an exe which is running only when work is actually being done? Boinc can detect a named exe running and pause. Boinc tasks have way longer deadlines and won't be affected by a folding task running occasionally. But running both at once means the folding could go too slow and run over the deadline.

[Peter_Hucker]

In case anyone wants to know, the one I needed is fahcore_22.exe, which I found in program data/FAH/ and so on. The name I got from what was using some CPU time in task manager - there may be others for different projects. I used the boinc manager to select the file from the folder tree and say "gpu only" so it paused the GPU.

So problem solved - Boinc runs when FAH doesn't have any work for Hawaii - computer number 7 of 10 is now doing science all the time on the GPU, with preference for folding - I see biology as more important than all boinc has left for GPU - astrophysics and maths.

[arisu]

I think WUs keep getting bigger and more demanding. So older hardware may age out sooner than one might expect when they can’t meet deadlines. Deadlines are not arbitrary. Research projects need to be completed in a reasonable time frame.

I can accept if my hardware doesn't meet "timeout", I'd stop using it. The loss of points is presumably to encourage users to stop using very slow cards, or ones which aren't on very long at a time.

But my "slow" cards are easily finishing on time.

Folding is inherently sequential, though there are multiple projects (and multiple copies of the same projects with slightly different starting parameters) which makes it easier to take advantage of parallel processing. Once you complete a WU and send it to the server, it is turned into a new WU for the next person. If your card is slow then it delays the entire "chain" for that project/clone/run combination). If you are currently on Project 12345, Run 7, Clone 0, Generation 42, then someone will only be issued Project 12345, Run 7, Clone 0, Generation 43 once you have finished (they start where you left off).

In theory, all projects could be split up into multiple speeds dynamically to maximize resource usage on any device, but no one has written the code for that (the logistics is probably not as easy as it seems) and the software used for the projects (GROMACS for CPUs and OpenMM for GPUs) aren't compatible, so you can't just run Project 12345, Run 7, Clone 0, Generation 42 on a GROMACS core and then create Project 12345, Run 7, Clone 0, Generation 43 for an OpenMM core. In theory it is possible, but in practice it would need a huge amount of development effort and even more testing effort.

I've read discussions about "streaming" results to the server instead of processing one big WU at a time. That could make it easier for low-end devices to help, but again it's a question about development time, and programmers and testers aren't free. In the end, there are more pressing matters to worry about. Low-end hardware is valuable to the project, but inefficiencies in the low-end aren't as detrimental as (existing) inefficiencies in the high-end.

tl;dr It's solvable but it's a matter of logistics and high development time and effort for relatively low payoff.

[Peter_Hucker]

Anyway, what I'd like changed is this error message because it's incorrect/misleading:

21:56:49:E :OUT430:HTTP_SERVICE_UNAVAILABLE: {"error":"No appropriate assignment"}

The second part makes sense, but the first part makes me think a server is broken. Why is it saying this?

[calxalot]

What http code would you suggest instead?

[Peter_Hucker]

Something which says what it means. I see service unavailable on Tesco where I order food home deliveries. It means their server is down.

Why would I be getting an http error here anyway? There is no error. It should be just a message saying no work surely?

I looked up 430..... From https://http.dev/430

"HTTP response status code 430 is an unofficial client error specific to Shopify and is returned by the server to indicate too many HTTP requests are being made using the API within a certain period."

"When the 430 status code is received, the server is refusing to process the HTTP request because too many HTTP requests are made to the API in too close a succession."

So the error means I'm hammering the server, which I'm not.

If for some strange reason you need an http code to tell my client there's no work, pick from the hundreds here, which a quick scan through shows many more sensible ones: https://en.wikipedia.org/wiki/List_of_HTTP_status_codes

[arisu]

It's not a 430 error, it's a 503 error. OUT430 is the thread prefix and it's how the log tells you it is a connection from the 430th request thread.

See here in the FAH client code where the HTTP_SERVICE_UNAVAILABLE status is handled, and the cbang library where the log prefix is defined.

I think it is already an appropriate response code. The other alternative would be 204 No Content, but that would prohibit the server from legally including a response body. It would also prevent FAH from implementing the Retry-After header in the future, which might be a good idea so the server gets to control how often requests occur, instead of relying on the hardcoded exponential back-off.

The 503 error means that the server received and understood your request, but has nothing to give you. The actual reason for that is given in the response body, i.e. "No appropriate assignment". The Tesco website probably puts something like "Under maintenance" in the body, if it even sends one.

If you don't want to see the thread prefix in the log anymore, create a config.xml with this content:

Code: Select all

<config>
  <log-thread-prefix v='false'/>
</config>