Page 2 of 2

Re: Nice and renice issue

Posted: Fri Sep 25, 2020 3:19 pm
by ronnieross
Seems to be an interesting point of view on Computer Science: you can't slow down a process because of it executes forever and ever at the maximum speed CPU (aside voltage and clock configuration) can run.

Re: Nice and renice issue

Posted: Fri Sep 25, 2020 4:02 pm
by Joe_H
ronnieross wrote:Seems to be an interesting point of view on Computer Science: you can't slow down a process because of it executes forever and ever at the maximum speed CPU (aside voltage and clock configuration) can run.
Well, you can, but that requires programming in Wait states or other non-productive code. But basically F@h falls into the "compute-bound" type of process, it has calculations always ready to be done with limited amounts of other types of processing that would cause it to wait on. The client is set to run it at the lowest priority, the OS scheduler will preempt processing F@h for higher priority process that run.

Re: Nice and renice issue

Posted: Fri Sep 25, 2020 10:35 pm
by ronnieross
Joe_H wrote:
Well, you can, but that requires programming in Wait states or other non-productive code. But basically F@h falls into the "compute-bound" type of process, it has calculations always ready to be done with limited amounts of other types of processing that would cause it to wait on. The client is set to run it at the lowest priority, the OS scheduler will preempt processing F@h for higher priority process that run.
I've found an utility, called cpulimit, that does the "dirty job" (maybe adding wait states or so...)

# sudo cpulimit -p PID -l 300

where PID is referred to the running FahCore_a8

It works.

Re: Nice and renice issue

Posted: Sat Sep 26, 2020 5:41 am
by gunnarre
How does does the cpulimit command affect efficiency? If it causes a lopsided effect on folding speed, due to poor thread synchronizaion, then it might not be a good idea to use.

Re: Nice and renice issue

Posted: Sat Sep 26, 2020 8:07 am
by ronnieross
gunnarre wrote:How does does the cpulimit command affect efficiency? If it causes a lopsided effect on folding speed, due to poor thread synchronizaion, then it might not be a good idea to use.
Efficiency goes down: time for finishing job is circa X 3 (depending on final parameter you set from the command line).

But, this way, I believe you can let folding go on overnight without having CPU thermal issues.

(...)