WU Completion Times - Does This Check Out?

[RoundelMike]

Sorry to create another thread, but I want to make sure that my new garden is performing as it should be, both for the [H]orde and for Stanford. I've been reading DR_K13's post about calculating target folding timeframes, and I can't come up with an intelligible answer, probably due to my own mathematical incompetence.

Here's the scoop: I have a Celeron 800MHz running the p887_p53peptide887 WU, and other comparable computers running the same project. According to FAH LogStats, it's taking the 800MHz Celery about 2:20:00 per frame (over two hours), with a total WU run time of about ten days. Is that slow or am I on target? I want to catch any configuration errors early before I've run the garden for a couple months.

I'm running each of the 800MHz computers on Ubuntu, and if it is folding slow, what could be the reason?

 

[ICE_9]

I have a 700 P3 with 256 MB of ram running the p1140_RIBO_FSpeptide_EXT_nospring (GROMACS) (600 points) right now on fold server. Time per frame: 2:04:20 (2.895 pph). Start date: September 6 04:30:00. Estimated finish date: Wed Sep 14 22:17:26 2005. It is arround 8 days for me. I understand that this is a different protien, different points scale, different processor with a different cache size, but this the closest config I had to your config. You seem to me to be right on or just a little slow. Any other comments from others?

 

[RoundelMike]

I have a 700 P3 with 256 MB of ram running the p1140_RIBO_FSpeptide_EXT_nospring (GROMACS) (600 points) right now on fold server. Time per frame: 2:04:20 (2.895 pph). Start date: September 6 04:30:00. Estimated finish date: Wed Sep 14 22:17:26 2005. It is arround 8 days for me.

Your setup is very similar, except that all my computers have only 128MB of RAM instead of 256. I'm not sure exactly how RAM size affects folding, but I understand that it does make a difference. I probably make the situation worse by having the Gnome GUI running too...

 

[ICE_9]

The fold server clients all run command line style setup. They run on minimal ammout of space, while the entire system is stored in the RAM. Probably helps a fair ammount on the overhead. I have had good luck with the fold server. I won't kid you though, it can be tough to setup, and everyone wants different things from their setup. I have the log page show how my windows boxen are running also. I have the log page forwarded to the internet and can see my setup at work and reboot, restart, or shutdown boxen as necessary. It is an excellent setup and it is always a pleasure working with unhappy_mage!

In short, it works great.

 

[unhappy_mage]

I'm running a p887 and a p888 on a duallie 933 with 256 (so 128 each, sorta) for 64 ppd. Time per frame is 1:03:30 according to EM3.

 

[Tigerbiten]

I work it out as 142/1.6/0.8 = 110 hours or 4.6 days for the full protien to fold.

Check in the FAHlog.txt file that the SSE boost is engaged.
At the top of the file there should be a line similar to.....
"Extra SSE boost OK."

Without the boost you will fold at half speed.

Luck............

 

[RoundelMike]

I work it out as 142/1.6/0.8 = 110 hours or 4.6 days for the full protien to fold.

Check in the FAHlog.txt file that the SSE boost is engaged.
At the top of the file there should be a line similar to.....
"Extra SSE boost OK."

Without the boost you will fold at half speed.

Luck............

Heh...there's nothing about SSE boost in my FAHlog.txt files. I'm an idiot. Do you have to turn it on manually with the advanced options?

 

[Mohonri]

Heh...there's nothing about SSE boost in my FAHlog.txt files. I'm an idiot. Do you have to turn it on manually with the advanced options?

You use the -forceasm flag.

 

[Kvothe]

You should see something like this in the FAHlog.txt file:

[14:30:15] Preparing to commence simulation
[14:30:15] - Ensuring status. Please wait.
[14:31:15] - Assembly optimizations manually forced on.
[14:31:15] - Not checking prior termination.
[14:31:15] - Expanded 260742 -> 826210 (decompressed 316.8 percent)
[14:31:15] 
[14:31:15] Project: 1911 (Run 29, Clone 3, Gen 6)
[14:31:15] 
[14:31:15] Writing local files 
[14:31:16][COLOR=Lime] Extra SSE2/SSE3 boost OK.[/COLOR]

Not all of the cores take advantage of SSE boost though I believe, so you could possibly be using one of those cores. You can force it on at all times though through the -forceasm switch, as sometimes SSE won't work if the client closes cleanly.

 

[RoundelMike]

Thanks for the help guys. I've just applied the -foceasm flag to my two computers here at the office and we'll see what happens, although even with the switch I still don't see an SSE message in the log file. The cores are tinkers though, and they may not have an SSE boost option. I'll monitor everything and post back here with results just for reference.

 

[RoundelMike]

Okay, after using the -forceasm flag with my computers they've cut their folding time in half. Thank you for the heads up.

For future reference, when I first ran the F@H client on the computers, it did have the SSE boost running. For one reason or another I ended up rebooting each of the computers, and when I restarted the client the SSE boost was off, as if it had defaulted to off for some reason. You may want to check your boxen if you have to reboot to make sure the boost is still on and hasn't turned off for some reason. We could pick up a few extra points that way...

 

[Kvothe]

That's how SSE boost works. By default it is on, but if the client shuts down improperly, on the next restart it will default to off. It's annoying behavior, but it is the reason that we suggest that -forceasm is active in all instances where F@H is running.