Some lessons learned in the last month, that are not in the FAH official manuals:
1) It's not just 16(+) cores to fold big. Depending on the CPUs, you won't necessarily hit the bonus points with 16 cores, and might not hit it with 24 cores. BEFORE buying stuff, ask around. It doesn't hurt to try if you have an existing 16+ core machine, but if you buy the wrong stuff, it won't git-er-dun. A very rough guideline is cores x ghz = 60. Some stuff as low as 40 will make it, but other stuff at 60 won't. So there is a wide range.
2) You REALLY need to run the Musky Install: http://hardforum.com/showthread.php?p=1037125470#post1037125470 . Even if you aren't familiar with Linux, just follow by Cut/Paste each step in a Terminal window. If you decide to use the 7.x GUI client, you will go slower, and there is less tech support available for folding big.
3) The Work Unit that is the barrier is the P8101. Just because you fire your machine up and it makes the bonus deadline with other WU's, doesn't mean it will fold big reliably. You must fold the P8101 faster than 34:30 per 1% (TPF). There is a huge difference between the P8101 and other WU's.
4) There a known problem with file handling that can stall a machine over 1 hr at the end of the job. The Musky Install fixes it. For that reason alone, you should run that way.
5) Don't worry if you get a SMP job or two or three. Your machine is probably just fine. They run out of BigAdv now and then.
6) The 6.34 client won't see more than 32 cores unless told to. Use the -smp 48 flag if you have 48 cores, or whatever number you need.
7) Another known problem is sometimes a good running machine will run "slow" when it starts a job. It's rare but it's a huge hit in performance, it can be 20% or more. If you just quit (CTRL-C) and restart and it goes away. But first do this to help troubleshoot the issue:
1) It's not just 16(+) cores to fold big. Depending on the CPUs, you won't necessarily hit the bonus points with 16 cores, and might not hit it with 24 cores. BEFORE buying stuff, ask around. It doesn't hurt to try if you have an existing 16+ core machine, but if you buy the wrong stuff, it won't git-er-dun. A very rough guideline is cores x ghz = 60. Some stuff as low as 40 will make it, but other stuff at 60 won't. So there is a wide range.
2) You REALLY need to run the Musky Install: http://hardforum.com/showthread.php?p=1037125470#post1037125470 . Even if you aren't familiar with Linux, just follow by Cut/Paste each step in a Terminal window. If you decide to use the 7.x GUI client, you will go slower, and there is less tech support available for folding big.
3) The Work Unit that is the barrier is the P8101. Just because you fire your machine up and it makes the bonus deadline with other WU's, doesn't mean it will fold big reliably. You must fold the P8101 faster than 34:30 per 1% (TPF). There is a huge difference between the P8101 and other WU's.
4) There a known problem with file handling that can stall a machine over 1 hr at the end of the job. The Musky Install fixes it. For that reason alone, you should run that way.
5) Don't worry if you get a SMP job or two or three. Your machine is probably just fine. They run out of BigAdv now and then.
6) The 6.34 client won't see more than 32 cores unless told to. Use the -smp 48 flag if you have 48 cores, or whatever number you need.
7) Another known problem is sometimes a good running machine will run "slow" when it starts a job. It's rare but it's a huge hit in performance, it can be 20% or more. If you just quit (CTRL-C) and restart and it goes away. But first do this to help troubleshoot the issue:
Whenever such issue is manifesting on 6.34 machines, please run fahdiag, pipe it into
pastebinit and paste resulting URL:
fahdiag setup:
Code:cd $HOME wget http://darkswarm.org/horde/fahdiag/fahdiag.sh chmod +x fahdiag.sh
running fahdiag:
Code:./fahdiag.sh | pastebinit