Page 2 of 2

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 5:59 pm
by TheMikeyDK
Devildoll wrote:As i interpret these stats...
The servers are handing out 70K WUs per hour. Which is a bump from 55K earlier this week, ( no clue what number was prior to coronatimes )
So there is a lot of work being assigned, but contrary to what the error message in the client is saying ( WARNING:WU00:FS01:Failed to get assignment from 'xx.xxx.xxx.xxx:80': No WUs available for this configuration )

There are WUs available, it's just that there are more clients asking for them than the servers currently have capacity to talk to.

That's just my guess, i have no real insight into the matters and could be wrong.
The question isn't about what is going on right now. The question is about if there are balancing/priorities depending on different hardware.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 6:05 pm
by Harmin
interperting what bruce said, we don't know and we can't know

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 6:33 pm
by Devildoll
TheMikeyDK wrote:
Devildoll wrote:As i interpret these stats...
The servers are handing out 70K WUs per hour. Which is a bump from 55K earlier this week, ( no clue what number was prior to coronatimes )
So there is a lot of work being assigned, but contrary to what the error message in the client is saying ( WARNING:WU00:FS01:Failed to get assignment from 'xx.xxx.xxx.xxx:80': No WUs available for this configuration )

There are WUs available, it's just that there are more clients asking for them than the servers currently have capacity to talk to.

That's just my guess, i have no real insight into the matters and could be wrong.
The question isn't about what is going on right now. The question is about if there are balancing/priorities depending on different hardware.

Well, anecdotally...
When i started my computer ( folding with rtx 2070 ) today, it had to retry for a long time until it got the first WU, but after that it has gotten 3 in a row ( albeit with a few connection errors in between, but still being able to download the new WU before the previous one had finished its last 1% ) . Seeming as if you get your foot in the door, you're in, unless you hit pause.

It would be interesting to see what happened to your server machine if you let it finish all the jobs and then pause for a while, and then start looking for new work, if that works and it gets handed work before your main machine, that might mean that there is some kind of hardware bias.

But that might not be worth the risk of ending up empty handed for hours.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 6:35 pm
by JimboPalmer
[I am just a user like you. Just very curious! None of this is official, but I have reason to think it is how F@H works]
TheMikeyDK wrote:The question is about if there are balancing/priorities depending on different hardware.
Kinda.

AMD has recently introduced New Graphics cards based on a family called RDNA. RDNA does not run Core_21 WUs, it must be aimed at Core_22 WUs,

The buzzword is Species number. RDNA is Species 7 for AMD. Before that was GCN and before that Terascale.
https://en.wikipedia.org/wiki/RDNA_(microarchitecture)
https://en.wikipedia.org/wiki/Graphics_Core_Next
https://en.wikipedia.org/wiki/TeraScale ... hitecture)

Nvidia also has several generations of Hardware but the last one that does not work was Tesla, the oldest one that DOES work is Fermi, and the newest one is Turing,
https://en.wikipedia.org/wiki/Tesla_(microarchitecture)
https://en.wikipedia.org/wiki/Fermi_(microarchitecture)
https://en.wikipedia.org/wiki/Turing_(m ... hitecture)

No one admits this officially but frequently there is a number after the Graphics card name; this number hints the performance of the card run 24/7. This helps grade cards so they get WUs they will complete before deadlines.

(an aside, any one card needs to support OpenCL 1.2 and Double Precision floating point math, AMD withheld DP math from laptops for a long time. The Helpers here get frustrated with AMD laptops as a card with the same name may or may not fold)
https://en.wikipedia.org/wiki/OpenCL
https://en.wikipedia.org/wiki/Double-pr ... int_format

Currently all CPUs that can fold at all fold with Core_a7 (It needs to support SSE2.0, so any Intel Pentium 4 any back to 2000, and AMD back to 2003) internally however, a7 chooses to run using avx-256 if available. (Intel introduced AVX in 2011, AMD followed in 2012)
https://en.wikipedia.org/wiki/SSE2
https://en.wikipedia.org/wiki/Advanced_ ... Extensions

I do not believe SSE or AVX is sent back to the server to choose projects, just number of CPUs.

So GPUs negotiate Species, while CPUs go by number of CPUs. I think.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 6:42 pm
by bruce
TheMikeyDK wrote:The question is about if there are balancing/priorities depending on different hardware.
Yes, and No.

If the servers are too busy to respond to your request, it's also too busy to evaluate your hardware's description. That evaluation happens when the server receives the information about your hardware and it is given the opportunity to choose which project to assign to you.

When the machine at my doctor's office answers the phone, the initial message says "If you're having a life-threatening emergency, hang up and call 911 or ...." but it has to answer the phone before it tells you that.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 7:07 pm
by TheMikeyDK
bruce wrote:If the servers are too busy to respond to your request, it's also too busy to evaluate your hardware's description. That evaluation happens when the server receives the information about your hardware and it is given the opportunity to choose which project to assign to you.
But that is exactly what I am not asking about, thats why I mentioned being aware of the increased interest of helping.

My question is based on a perfect world where there is no such thing as server load, speed of transfer, and so on.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 7:09 pm
by bruce
Right. The "problem" goes away in your perfect world when the server has time to process the request.
i.e.: yes and no.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 7:15 pm
by TheMikeyDK
bruce wrote:Right. The "problem" goes away in your perfect world when the server has time to process the request.
i.e.: yes and no.
There was never a problem, there was a question if there was balancing/priorities depending on hardware.

An example, we got a 4 thread cpu and a 8 thread cpu, both idle. We got two WU's requiring 4 threads.

Will the 4 thread cpu get both one after the other, or will the one with 8 thread capability get the other, as it got nothing to do anyways?

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 7:23 pm
by Jesse_V
TheMikeyDK wrote:
bruce wrote:Right. The "problem" goes away in your perfect world when the server has time to process the request.
i.e.: yes and no.
There was never a problem, there was a question if there was balancing/priorities depending on hardware.

An example, we got a 4 thread cpu and a 8 thread cpu, both idle. We got two WU's requiring 4 threads.

Will the 4 thread cpu get both one after the other, or will the one with 8 thread capability get the other, as it got nothing to do anyways?
The client software is configured to use a certain number of CPU cores. For example, I have 8 cores on my CPU, but I've configured the software to only use 4 cores because I need the other 4 for other purposes. With this setup, the workunits will only use 4 cores, just as I set up in the configuration. This will be the same for you. Do you have the client set to full power?

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 7:26 pm
by TheMikeyDK
Jesse_V wrote:The client software is configured to use a certain number of CPU cores. For example, I have 8 cores on my CPU, but I've configured the software to only use 4 cores because I need the other 4 for other purposes. With this setup, the workunits will only use 4 cores, just as I set up in the configuration. This will be the same for you. Do you have the client set to full power?
I don't think you understood the question. This isn't about configuration on the client side, it is about what is going on at the server side.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 8:00 pm
by JimboPalmer
TheMikeyDK wrote:I don't think you understood the question. This isn't about configuration on the client side, it is about what is going on at the server side.
None of us own a F@H server, neither an Assignment server like you are asking about, a Work Server, which actually downloads the WU, or a Stats Server.

The best we can do is guess. I did.

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 10:34 pm
by davidcoton
From observation rather than code knowledge, the WUs are assigned to suitable hardware based on project priorities. There is no attempt to maximise the WU size that the hardware can take. This will happen naturally to some extent as lower power PCs are allocated the smaller WUs, leaving the larger ones for the more capable PCs. Of course a powerful PC will fold a small unit very quickly, and come back for more, so there are not many situations in which folding power is wasted. Those I have seen semm to be due to limitations in the GPU drivers -- which are updated outside F@H's control (though F@H does communicate with the larger GPU manufacturers).

Re: Trying to understand WU distribution.

Posted: Thu Mar 19, 2020 10:56 pm
by TheMikeyDK
davidcoton wrote:Of course a powerful PC will fold a small unit very quickly, and come back for more, so there are not many situations in which folding power is wasted.
Now when you mention it, a few hours before the servers ran out of work, my main computer did start to get some pretty small WU's which only took 15-20 minutes to complete. That could indicate the ones I usually get were gone, and it had nothing else to send, so it started to assign smaller WU's.

Re: Trying to understand WU distribution.

Posted: Fri Mar 20, 2020 10:31 am
by davidcoton
AIUI at present Coronavirus projects are at high priority, which obviates the need for a COVID19 "cause" setting. So when they are available (most of the time now) and when the server has a moment to serve your request, you should get Coronavirus work, whatever the size of the unit.