PDA

View Full Version : Running f@h on HT systems



Mustard
12-26-2004, 02:32 PM
Just a question here.... some projects go great on HT systems running linux, like seti classic for instance while others are better off with just running one instance only. What is best for f@h? I haven't done any timing runs on it since it takes a while to complete the work units.

My system has 512 mem, and is a 2.4c HT. Would this be better off going for the BIG units that need lots of memory, or staying with the normal gromacs units that have expirations dates? The system is running linux.

magnav0x
12-26-2004, 06:50 PM
Sorry I don't have a answer for your HT question, but I'm also curious if it's better to run work units with expiration dates or run the smaller ones that don't have expiration dates. Most of the computers I have on the project are "decent", but aren't on 24/7....they are on maybe 40-50 hours a week.

If you run a P4 2.8 (just an example) on units with no expiration will you get roughtly the same weekly output as you would if you ran larger units with expiration dates.....or is there some sort of bonus for running larger units? I know the smaller units are like 30-40 points and they take 12-24 hours to run any how.

Mustard
12-26-2004, 07:31 PM
Well I've got one long one running on the system, so will see how it goes. Then I'll give it a stab with two of the same work units and see how that goes. the only thing is to try and end up with a couple of the same units. At least under linux it is easy to move stuff around without gumming up the works. :)

IronBits
12-26-2004, 07:56 PM
You sure can...

http://folding.stanford.edu/console-userguide.html
* Each different running copy of F@H has to have its own Machine ID number. If you download each copy of F@H from the web site and configure fresh, there will be no problems.
* Please don't manually retry to send WUs soon after a netsend failure. The netsend failure means that the server was probably overloaded and retrying will just make things worse.
* Do not run multiple copies of FAH from the same directory

You will need to create 4 directories, give each client it's own working directory.
Run each one manually the first time and give each one a different ID.
Write a quick script
cd [FAHdir1]>fah -local & > /dev/null
cd ../[FAHdir2]>fah -local & > /dev/null
cd ../[FAHdir3]>fah -local & > /dev/null
cd ../[FAHdir4]>fah -local & > /dev/null
From then on, just run the script :D

Mustard
12-26-2004, 08:24 PM
oh, I know I *can* run more than one instance, what I was wondering is is there any performance gain out of it, it is it better and faster to just run one instance when viewed totally from the point of gaining maximum points over a period of time?????

IronBits
12-26-2004, 08:34 PM
Surely two is better than one :smoking:

maefly
12-26-2004, 09:10 PM
Read this:

http://forum.folding-community.org/viewtopic.php?t=10427

Mustard
12-26-2004, 09:24 PM
Originally posted by maefly
Read this:

http://forum.folding-community.org/viewtopic.php?t=10427

Thank you very much for the URL. An interesting read! :)

IronBits
12-26-2004, 09:40 PM
:bang: :bang: :bang:
1 client per physical processor
:blush:

Thanks for clearing that up! :cheers:

willy1
12-26-2004, 09:55 PM
There's gaping holes in that logic.

First it assumes all client CPUs are the same speed, and we know that's not true. If I can finish 2 HT WUs in less time than the guy doing one WU crunching along on his P3-800, who's holding up the processing?

I don't see the 1 day = 300 day thing at all. It doesn't even address parallel processing by multiple clients/users.

Beside, stats are everything :jester:

dragongoddess
12-27-2004, 01:34 AM
you get a 10% to 30% gain by running two clients. The type of wu does not seem to matter. The timeless wu's are really for those slower machines. If one has a 2.8ghz machine they should not be running the timeless wu's. You are in effect taking wu's away from those who have machines that cannot finish a wu by its deadline date.