I just folded my first protein

[bs_texas]

So, I was contemplating the end of the universe today and then, you know, if there is an end to the universe then what's stopping it, and if there is something stopping it, then, you know, what's behind what's stopping it and ... *

No wait, that was the other day, today I was contemplating Points vs. Work Units. I know I could probably research this, and it may have been answered back up there somewhere in this 600 some odd posts (heck, I may have even already asked this before), but I'm curious about the -value- of points vs. the -value- of work units to the folding process. I assume points is fun to watch and compare, but I would think that completed work units is the thing.

500,000 points looks good, but next to only 2200 work units.... hmmmm...

Comments?





(* from Spinal Tap, of course)

[happybunny]

Oh no!

I've lost my coveted 5th place spot which I held for so long.
http://fah-web.stanford.edu/teamstats/team36480.html

Just bumping this post for anyone who hasn't seen it lately.

[cybertron]

I guess folding likes my new quad core.

[je_fro]

/me grumbles....
I've been moving and changing jobs...

[bs_texas]

Looks like I've grabbed that coveted 5th spot. But, it might not last long with cybertron hot on my heels.

Cheer up je_fro. I moved not long ago but I got set back up in short order.

As a team, I've watched us move from about 147th place (as I recall) up to about 180th place. It's the wrong direction, but at least other teams are moving up and the work units are getting done for the folding project.

Adding 3 64bit machines to my arsenal has helped, the rest are older machines and my connection to the internet is 56K dial-up.

[bs_texas]

Hmmm... I don't know exxactly what happened, but my 2 dual core 64 bit vista machines were running fah in smp mode just fine (I think) then, it seems, all of a sudden I can't send in my results because the .dat files are over 20MB big! And I'm on a 56k dial up behind a proxy that disconnectes me every 4 hours!

I think I read that the smp option takes the big work units and there's no real option for smaller work units. I'd like to be able to punch 'em out quick and small, but apparently that's not happening anymore.

My ubuntu machine is 64 bit, but not dual core.


Anybody have some setup tips for this?

[techwise]

There has to be a way to hack out a solution. What about splitting that file, sending the pieces to a remote server with a bigger pipe one at a time and then joining them on the remote and sending them in from there?

It should even be scriptable. Cant you elect not to send completed units and continue to crunch units? This would keep you from having downtime on crunching and allow you to return the units at your speed, presumably as long as its prior to the deadlines.

[cybertron]

bs_texas or anyone else using the smp client: Does it use anywhere near all of your CPU? Mine is currently using < 40% of my CPU. On a good day it maybe uses 70-80%, but it never uses 100% like I think it should. I may just go back to running four instances of the standard client since it would definitely get more work done.

[je_fro]

I noticed the same thing with my new smp client, but after awhile it went back to using all the cpu. Perhaps it's just the WU's you're getting?

[techwise]

bs_texas or anyone else using the smp client: Does it use anywhere near all of your CPU? Mine is currently using < 40% of my CPU. On a good day it maybe uses 70-80%, but it never uses 100% like I think it should. I may just go back to running four instances of the standard client since it would definitely get more work done.

How many procs/cores do you have? I remember FAH recommending to not use "-smp" on dual core machines. I think they recommend four or more for smp. Here is how mine look to top using 2 separate instances running in smp mode on a dual proc 8 core system. It has always run in this way for me.

PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
5367 michael 39 19 81584 40m 1880 S 99 1.0 636:11.29 FahCore_a1.exe
6718 michael 39 19 47984 16m 1880 S 97 0.4 56:55.73 FahCore_a1.exe
5369 michael 39 19 77748 37m 1664 S 78 1.0 495:37.78 FahCore_a1.exe
5368 michael 39 19 77616 37m 1664 S 78 1.0 508:52.62 FahCore_a1.exe
5370 michael 39 19 78212 38m 1676 S 78 1.0 484:00.73 FahCore_a1.exe
6719 michael 39 19 46052 14m 1664 S 54 0.4 30:59.13 FahCore_a1.exe
6721 michael 39 19 46456 14m 1676 S 52 0.4 30:42.92 FahCore_a1.exe
6720 michael 39 19 45928 14m 1664 S 48 0.4 28:54.64 FahCore_a1.exe

[bs_texas]

Dang it!! Well, I'm out on another TDY and I haven't had time to go back and figure out the smp thing. But, the 'dang it' is about being away and finding out there is a problem with the phone system back home. I probably won't be uploading anything back to F@H until I get home at the end of September or beginning of October!!

I'm on a 56K dial-up system and my smoothwall does a persistent re-dial when it loses connection, but I threw in an arbitrary 30 dial limit a long time ago thinking that would be sufficient. I'm pretty sure that has been reached and all 7 of my computers there are just sitting idle, going... 'cannot send results' and 'cannot connected to assignment server'!!

Fortunately, my laptop at the hotel is still folding!

[cybertron]

Oops, I missed the previous responses here. I wonder why I didn't get an e-mail notification?

Anyway, I don't think it's just my work units because it's been happening consistently since I got my quad core in April. techwise's top output appears to be about what I'm seeing too, where each instance has one process using 100% of a core, and the rest are using varying amounts less (though I'm only running a single instance).

I see on the Folding forums that the a1 core I seem to be using does not scale very well on four cores and it also seems to be pretty old (2006). I've enabled the advanced methods to hopefully get a newer one since even deleting the core doesn't seem to cause it to get a new one right now. I guess I'll have to keep an eye on it.

[techwise]

Oops, I missed the previous responses here. I wonder why I didn't get an e-mail notification?

Anyway, I don't think it's just my work units because it's been happening consistently since I got my quad core in April. techwise's top output appears to be about what I'm seeing too, where each instance has one process using 100% of a core, and the rest are using varying amounts less (though I'm only running a single instance).

I see on the Folding forums that the a1 core I seem to be using does not scale very well on four cores and it also seems to be pretty old (2006). I've enabled the advanced methods to hopefully get a newer one since even deleting the core doesn't seem to cause it to get a new one right now. I guess I'll have to keep an eye on it.

Im still getting a1 cores as well. This must be the desire of the project at this time. Thats all I can figure. However on my 8 core systems I am getting lots of work done so if the newer cores are better then they'll really be cranking out the units.

[cybertron]

Yeah, I've heard good things from people using the latest a2 cores, so I'm hoping I can get one of those now. From further reading on the forums, you do have to have the advmethods switch enabled either on the command-line or through the config file in order to get a2 cores. They're also not getting assigned to >2 core machines much right now for some reason, which I suppose is partially because they're the only ones that are really useful for SMP on dual core.

I also read something about running the Windows GPU client through Wine, so that might warrant some more investigation...

[techwise]

Just a quick update. On one of my 8 core boxes I just noticed that I have 2 units folding, both units using the a2 core and all 8 cores are running at +90%, not scaled like they were on the a1 core jobs.

Cheers

Mike