Folding@Home Weekly Team News - 31st March 2006

[GT3]

1200 / 7 = 171.4 ppd. That sounds very good to me.

Really? I'm used to seti units being completed in a couple of hours, the folding units take a couple of days!

But it is heavily overclocked, 2.4ghz at 220fsb 1:1 from a little XP1700

 

[BillytheImpaler]

According to the way point values are calculated, the overall global average should be somewhere around 110 ppd. To exceed that, especially on an older machine, is a great sign.

 

[Ricko]

Well, finished work for the day. Got 3 servers of the above spec Folding for OcuK now!

Will be interesting to see what sort of results they produce over the weekend. Especially the Linux one, as it seemed to be running about 8 folding processes but only 4 were using any cpu time

Rick.

 

[BillytheImpaler]

...as it seemed to be running about 8 folding processes but only 4 were using any cpu time

My Linux boxes do that too. I don't quite know why as I'm too much of a *NIX n00b. Looks like so when I run

ps -e

There'll be one process per core that has CPU time associated with it and a slew of other core processes that haven't used a second of CPU time. Perhaps somebody else knows why this is but it all seems to work so I'm not concerned. I suspect it has something to do with the way the kernel wants to schedule the processes.

 

[Ricko]

Thanks for the info BillytheImpaler. I was concerned as I would be in deep trouble if I knakered the server!

But if it is normal I will ssh in remotley tonight and install F@H on the other 4 Linux servers, of the same spec as above

Cheers

Rick.

 

[BillytheImpaler]

How are you installing it on them, anyway? Are you using ivoshee's finstall script? I'm just curious.

 

[Ricko]

Yeah, I used the script from the sticky. Worked well, the servers are running Suse 9.3.
When running the ./folding start command it doesn't work though. So I have to go into each CPU folder and run the ./FaH command. This seems to work ok, but I have to laeve the shell window open

I really don't know enough about linux at the moment, but it does not seem to be causing any problems.

Rick.

 

[BillytheImpaler]

Shell window? What's that?

As these are active servers I imagine that you ahven't tried a reboot. Is the folding script there? It should be (if you followed the instructions exactly) in /home/<username>/foldingathome. What happens when you run

ls

when you're in foldingathome? Does the file called "folding" exist there?

 

[Ricko]

Hehe, a command line man eh? The servers need to have X running as they are terminal servers, thats my excuse anyway

Not tried a reboot yet, will give that a whirl on Monday.
The foldingathome folder is in /root which may be why I get 4 'access denied' errors when running ./folding start. The folding file is in /root/foldingathome/

Rick.

 

[rich99million]

In future would it be possible to have a top list of only those using a single machine? Like the old seti stats?

Just a thought.

yes I don't see why not - with only around 100 active team members it shouldn't be too taxing

as you've already said we would really need to keep a list for those with only one cruncher - an automated system would be great but for the reasons Billy stated that's not gonna happen
i'll keep the news the same for today but will get working on a nice looking format for it next week - i'll post a thread at some point asking for single machine crunchers to register

the only problem i can see is how we will class a "single machine" - I would say a P4 with HT is a single machine but how about an X2 or a dual Xeon?
I guess it's just down to individuals to decide wether their machine qualifies to be counted as one rather than many - that's fine by me

edit: oh and way to go on the XP1700 - I have two of the little beauties crunching for me, fastest is 2200MHz though I'm afraid that as an overclocker I basically suck

 

[BillytheImpaler]

Yeah Rich, if you're up for keeping track of single-users in that way the best way to go about it would ahve a sign-up. Folks post in next week's news (today ) and ask to be added to the single-machine watch list. It's on the honor system for the most part for people to distinguish a single machine from a pair or more. That seems like it'd be the most equitable way to do it.

Rick, I know that in the finstall page at the official FAH wiki, they warn agaisnt installing whilst root. They never quite say why though. Perhaps you've discovered the reason? If you're going to reboot the machine next week you might want to uninstall the installation in /root and put it in a regular user's home directory. If that fixes it we know what the problem was. I imagine that it's simply an issue with the way the scripts are written but I'm not bold enough myself to start poking around in them.

I might play with it when I get home from work.

 

[Ricko]

Hmm, you are probably right.

As far as I can tell though, the only local user is root. It gets the rest of the users via LDAP. The /home folder is empty, each user's home folder is mapped from a NAS server via NFS to /homes.

Dunno I'll just have to have a think

Rick.

 

[BillytheImpaler]

Just for the sake of experimentation, how about you create a new, local, non-privileged user?

/goes to look up the command to do that...

EDIT: http://linux.about.com/od/commands/l/blcmdl8_useradd.htm
That tells all the vital details of the useradd command. If only I could decipher it...

 

[Ricko]

Just for the sake of experimentation, how about you create a new, local, non-privileged user?

Yup, was thinking that myself.

And it works!
The command by the way is 'yast' then you fill in some boxes and go next a few times

So make that 2 Windows servers and 5 Linux servers, with a total of 14 3.06Ghz xeons and 28 folding threads.

Cheers for you help!

Rick.

 

[BillytheImpaler]

Huzzah!

Those'll be folding monsters once QMDs come back. Though you might have to reconfigure then if the incredibly high memory useage of the QMDs are too much for them. I'd hate to see you bog down business for some silly DC app.

Motto: WWRD, what would rich99million do?

 

[Ricko]

Cool, what are QMD's?
Unfortunatly the servers are very memory dependant, they often end up with only a couple of hundred mb free from 4Gb when they get busy!

Edited above post

Rick.

 

[rich99million]

So make that 2 Windows servers and 5 Linux servers, with a total of 14 3.06Ghz xeons and 28 folding threads.

wowsers
very nicely done - and a great bit of teamwork figuring that one out so quickly

Motto: WWRD, what would rich99million do?

hehe if only i knew anything about Linux or had access to that kind of power
the place where i work is run on hamster-powered BBC Master Systems, we don't even have a server as everything is run remotely accross something i would imagine is not far removed from a 56K modem connection to a server halfway across the country

also you won't be singing my praises when you realise I missed the usual stats save point so this week will be an hour longer than normal
that'll teach me to start a conversation about how rubbish our work place is with the missus on a friday

 

[BillytheImpaler]

QMDs are work units that use the QMD (Quantum Molecular Dynamics) core, Fahcore_96.exe. Currently the most computer intensive core FAH uses.

From Stanford's QMD FAQ: For other molecular dynamics simulation packages, interactions between atoms are described by an empirical potential energy (“force field approach”), which is an analytical function of atomic coordinates. In QMD core, there is no force field – atomic interaction is calculated using quantum chemical method or by solving Schrödinger equation.

For non-chemists we can say that it's one of the FAH team's newest and most advanced folding simulation. The units that use the QMD core, FahCore_96, are p1910, p1911, and p1912. They're each worth 450 points and take a day to complete at 100% utilization, give or take a few hours.

QMD units are highly sensitive to memory bandwidth. Since they use up to 350 MiB physical memory and another 400 MiB virtual memory before "converging", they can easily saturate the system bus in a PC. Since they use so many system resources QMDs are rewardeded bonus points over and above the calculated point value Stanford generated.

Currently ONLY INTEL NETBURST based computer systems can revieve QMDs. There are no exceptions. This is becasue of the EULA associated with the FORTRAN compiler and libraries used in creating the core. They were written by Intel and create code that refuses to run with optimizations in an AMD system. There are backdoors around this but it is safe to say that for the time being and the near future, QMDs are an Intel-only party.

At the moment the WUs are on hiatus, awating analysis from the researchers who create them. They have been so since the beginning of the year. the Pande group refuses to predict when their analysis might be complete but many are waiting with bated breath.

EDIT: I need to either re-read my posts or invest in a keyboard with working shift keys...

 

[Ricko]

Ah, I see, thanks for the info.

Might be able to get away with a WU like that, without it knackering the system. Will have to wait and see!

Rick.

 

[BillytheImpaler]

If it turns out that you can't get away with it, just go back through each configuration and change the "set -advmethods always" to no. To get QMDs you must both -advmethods and big WUs enabled.