Folding Forum

I have a 40 core server, which has successfully gotten assignments before, but is failing to get them now, (over 24 hours without one)...
Its gpu has successfully gotten one, but the CPU's are all idle!

Note I can get to the web pages fine at the addresses below, there is no firewall enabled (plus the gpu is working)
Any ideas?
Eric


12:57:40:WU00:FS00:Connecting to 171.67.108.45:8080
12:57:40:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.45:8080': Empty work server assignment
12:57:40:WU00:FS00:Connecting to 171.67.108.204:80
12:57:41:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.204:80': Empty work server assignment
12:57:41:ERROR:WU00:FS00:Exception: Could not get an assignment
13:01:54:WU00:FS00:Connecting to 171.67.108.45:8080
13:01:54:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.45:8080': Empty work server assignment
13:01:54:WU00:FS00:Connecting to 171.67.108.204:80
13:01:55:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.204:80': Empty work server assignment
13:01:55:ERROR:WU00:FS00:Exception: Could not get an assignment
13:08:45:WU00:FS00:Connecting to 171.67.108.45:8080
13:08:46:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.45:8080': Empty work server assignment
13:08:46:WU00:FS00:Connecting to 171.67.108.204:80
13:08:46:WARNING:WU00:FS00:Failed to get assignment from '171.67.108.204:80': Empty work server assignment
13:08:46:ERROR:WU00:FS00:Exception: Could not get an assignment

Some extra info: note the CPU's say 32, but there are 40...not sure why that is, pretty odd. This is from the Sysm Info tab:
Suggestion, make the system info easy to copy/paste, its impossible to select it now.

Version 7.4.4
SVN rev 4130
CPU intel cpu
CPUS 32 (there are 40)
Memory 36 gig
WINDOWS_Threads
windows 7 pro
os version: 6.1
GPUs 1
GPU0 Nvidia:2 gf119 [nvs 315]
cuda driver 6000
win32 service false

I'll wireshark the connection to the work server and get a dump of the packets sent/received, maybe they will have useful info.
E

Ok, I have a wireshark capture of the failed to get work, you can download it from this link

https://drive.google.com/open?id=0B6FBW ... UgxTGNvMnc

Is there a place to download the source code? I'd run it in a debugger and see if I can catch anything client-side...github?
Thanks
Eric

OK, it just got a work assignment, not sure why, nothings changed, it went 24+hours without one, checking every few minutes, but just managed to get one....so all good....for now.

Running Windows, my understanding is that 32 is the maximum number of cores that can be used at once. Beyond that, there has been continuing shortages of CPU WU's that can be assigned to more than 24 cores.

System information is easily posted by copying the first 100 or so lines of the log file. Those can be shown by using the Refresh button, or you can open the log file from the data directory in a text editor. The advantage of posting the beginning of the log file is that the folding configuration can also be included.

Source code for the client is not available.

If you want to use more cores than 32, you can configure additional CPU folding slots and set them to use the remaining cores. You do need to reserve at least one core for each GPU folding slot.

Thanks will try upping the cores, as a note, you might want to change the license for the client, its GPL v3.0 (about box), which means the source has to be made available...
E

Is that the About box for FAHClient, or FAHControl? FAHClient is not open source, FAHControl is. See this posting on the subject by the PG - http://folding.stanford.edu/home/faq/faq-opensource.

ahhh, I see, its the client control box, which is gpl...ok, the client contains and is not dependent upon any gpl code for execution then.

EricAlbers wrote:Thanks will try upping the cores, as a note, you might want to change the license for the client, its GPL v3.0 (about box), which means the source has to be made available...
E

FAHControl has been made available. https://github.com/FoldingAtHome/fah-control FAHClient isn't open source for security reasons, but is built using open sourced code (CPU - GROMACS, GPU - OpenMM).

Running 32 cores will probably perform as well any setting above that, especially if you plan to fold on the GPU as well.

Cool, it would be nice to have a 'complete' open source solution, I understand people were abusing the points system (hard to believe given this stuff will save lives, but people are weird)...

There are many science fields which could benefit from such a nice system as this, (particle physics for example) who have large datasets to work through...

Its a shame you couldn't calc the points ahead of time before the work units are sent out, then theres no ability to tamper with the scores and no incentive to try to get an advantage

Well the base points are calculated ahead of time, but since processing time will vary on different systems the bonus can not be. But that is only part of it. Attempts at score tampering were detected even before bonus points were involved.

Unfortunately, since as you say - "people are weird", not all the reasons for the closed source portions that validate that WU's are genuine, etc. are as benign as just abusing the points system. Discussing that would really take this post way off topic.

FYI, while my server got a work unit finished a few minutes ago, its once again unable to get more work,
failing to get work assignments from both 171.67.108.45 and 204


IDLE CORES ARE THE DEVILS WORKSHOP!!

That just means the two Assignment Servers can not find any Work Server with a WU that is assignable to a system with as many cores as yours. Some other folders with large core count servers have split the cores over a couple CPU slots, 24 and 12 might work well for your system. Those thread counts would avoid the problems that sometimes come from numbers with larger prime number factors.

OK, that seemed to work, I created 2 slots each with 20 cores, they both took off....slot 0 never did fill even when I set it to 20 however, so I just paused it, slot 1 is gpu, 2 and 3 each have 20 and are running now....

Thanks!
Will make 4 10 cpu slots if I need too.
E

One CPU core should always be reserved for each GPU that folds. The rest of the CPU cores can fold as needed. Just note that prime numbers of cores are not supported. 11, 13, 17, etc. The data doesn't split up and process well like that. Non-prime odd numbers are supported, most of the time. The preference is for even numbers of cores (except in large core numbers, 2x19 isn't much better than 1x19) Even numbers with multiple factors work the best. 12, 24, 36, etc. YMMV.