Reply With QuoteFor mult-processor machines. Probably should have been named "clients" 1 thru 4
OK, I've read Larry's FAQ over at the really cool looking TER website, and I've looked through our forums, but I still have a question...
What's with the Machine ID option during advanced client configuration? It gives me the option of setting it between 1 and 4. I don't think this is the unique client ID because that appears to be assigned by the F@H server on the first startup.
So, what is it?
For mult-processor machines. Probably should have been named "clients" 1 thru 4
-:Beyond:-
OH! rut-roh... /me runs off to check new installs....
Ah, thanks, Beyond. Since you're batting a thousand so far, how about another one?
How much difference is there speed-wise between the Linux and Windows clients. I've read conflicting reports on the subject...
TBH, it is hard to say. I have not been running the clients long enough to get a feel for the (if any) speed differences between OS's. I think it would take several weeks of close study to see any differences as there are so many different proteins being released at any given time.
-:Beyond:-
Be sure to run with the -advmethods switich on those P4's, there are some Gromac's proteins that take advantage of the P4's SSE code making them fast.Originally posted by Angus
OH! rut-roh... /me runs off to check new installs....
-:Beyond:-
P4s? What P4s? What is -nonet?
/me looks innocent
er- What is 'Gromacs' ? I saw two cores in the install choices - f@h and g@h . I selected f@h since that's what I thought you were all talking about.
That be itthere 3-4 cores depending on what proteins are available and your selected core choice, you will get Gromac's,Tinker, Genome workunits. The Gromac's are the newest, biggest, highest scoring workunits..not to mention the toughest on your hardware.
-:Beyond:-
This may be putting a damper on things until it is cleared up.
-:Beyond:-
Throwing the -forceasm at my F@H client on the Cellies has helped speed out some, I'll see how much it helps the P4 and Athalon XP later this evening.
A word to the wise....
If you are running a pharm and monitering the clients with EM, EM does not catch the error,client not producing work as stated in my second problempost on the FAH forums Check those clients!!!!
Oh one more thing...the clients I had (key word "had") running the "genome core" exclusively as a test have for some odd reason changed to the folding "tinker" core on thier own!
-:Beyond:-
-:Beyond:-
I'm running 2 boxen, both started late last night.
1. Athlon 2100+ under Win2K with Gromacs core. This one locked up middle o'night & had to be restarted this morning. It's now on "step 32", whatever that means.
2. Athlon 2200+ Linux box with Tinker core is at "frame 252". It ran all day, all night, Marianne.
These numbers sound way, way off or about right?
BTW, both are text-only clients - no fancy-schmancy stuff.
Last edited by Paratima; 05-02-2003 at 07:03 PM.
Sounds about right, taking into consideration the "shutdown" on the first one last night. They are more than likely '73' point workunits. My XP1700/1800's take about 28-29 hrs to complete one.
-:Beyond:-
Ah! Thanks, Beyond. I haven't run F@H since April of '02 & I've lost the feel for how fast it goes.
(Not to mention I've got better firepower nowadays.
If these two keep running, I'll toss in a couple more & see how it goes.
Last edited by Paratima; 05-02-2003 at 08:11 PM.
It is Alot slower now, but the wu's for the most part are larger and the score is better than ever.
-:Beyond:-
After some curiosity set in on the Athalon XP and P4 machines they got the feel for the -forceasm switch.
Without -forceasm and with -advmethods
p671_TZ2_EXT_EXP on P4 1.6 0:21:00 per frame
p671_TZ2_NAT_EXP on Athalon XP 1800 00:16:00 per frame
With -forceasm and with -advmethods
p671_TZ2_EXT_EXP on P4 1.6 01:12:21 per frame (retesting)
p671_TZ2_NAT_EXP on Athalon XP 1800 00:31:12 per frame (retesting)
Last edited by magnav0x; 05-02-2003 at 09:31 PM.
Please get back to us....if the forceasm is as good as it seems on the XP's then we may very well be able to go "flying" up the charts.
-:Beyond:-
Seems well EMIII still hasn't update, I'm thinking the real numbers will be quite a bit higher, but should be less than 20mins /me hopes.
*edit* I checked back and the numbers were way off on the Athalon XP with forceasm. I'll be sure to watch em longer before I say anything, but i'm gonna try out some more things. And retest it on the P4. I'll do some tests with -forceasm and no -advmethods, but I probably won't be able to sit and watch em long enough before I have to leave tonight. I gotta get down to the 1/8 tracks tonight and do some racing
Last edited by magnav0x; 05-02-2003 at 08:23 PM.
okOriginally posted by magnav0x
Seems well EMIII still hasn't update, I'm thinking the real numbers will be quite a bit higher, but should be less than 20mins /me hopes.
*edit* I checked back and the numbers were way off on the Athalon XP with forceasm. I'll be sure to watch em longer before I say anything, but i'm gonna try out some more things. And retest it on the P4
-:Beyond:-
Based on magnav0x's experience, I changed my one W2K Athlon 2100+ to -advmethods and -forceasm. I'm not using EM yet, so I just sits & watches the console.
It just did its first full update a few minutes ago & it looks like the frame time went from 37 minutes to (are you ready for this?)... 25 minutes!
As long as it doesn't increase the crashability, it sure looks good to me!
Dat be's what I'se yoosing, Jah!
I'll update this thread in an hour or so.
What exactly is the -local switch doing? I see it mentioned a lot but not sure what it does.
The -forceasm switch didn't do a great deal for the Celly 1.4's, but it did knock a minute or two off of the frame time, which for a 100 frame WU is still an extra 100-200 minutes
It tells F@H to stick all its activity into current folder - useful mainly on multiprocessor machines.
rm -Rf /
I'm going through my logs from when the machines were running with just -advmethods to see what the original numbers were, because I was just guestimating earlier
ruro! Didn't watch my log files good enough! I've been terminating F@H via task manager on the Win2k machines (all of the F@H machines lol) because I'm running it hidden with hideit.exe. Well......
--------clip from logs when staring the client up---------
01:16:58] Preparing to commence simulation
[01:16:58] - Ensuring status. Please wait.
[01:17:15] - Looking at optimizations...
[01:17:15] - Working with standard loops on this execution.
[01:17:15] - Previous termination of core was improper.
[01:17:15] - Going to use standard loops.
---------eof---------
No wonder they've been going so slow! Doh! I had to start em up and close em right
Without -forceasm and with -advmethods
p671_TZ2_EXT_EXP on P4 1.6 0:21:00 per frame (numbers with optimization enabled)
p671_TZ2_NAT_EXP on Athalon XP 1800 00:16:00 per frame (numbers with optimization enabled)
With -forceasm and with -advmethods
p671_TZ2_EXT_EXP on P4 1.6 01:12:21 per frame (retesting)
p671_TZ2_NAT_EXP on Athalon XP 1800 00:31:12 per frame (retesting)
I noticed that when I shut it down incorrectly and it went to using standard looping, the WU's were taking double, yes double, amount of time to do. Be sure to check on this guys! I'm retesting the clients with -forceasm with correct optimization. I'll have the new numbers for those when I get from the races, late tonight.....er about breakfast time Saturday morning
Last edited by magnav0x; 05-02-2003 at 09:36 PM.
I'm not seeing what I thought I saw earlier.
This here is gonna take a re-test or two.
(At least it hasn't froze up again!)
With -forceasm and with -advmethods
p671_TZ2_EXT_EXP on P4 1.6 00:22:00 per frame (with optimization)
p671_TZ2_NAT_EXP on Athalon XP 1800 00:16:12 per frame (with optimization)
So after letting them both sit in a long time with -advmethods -forceasm switches they acutaly seem almost identical as without -forceasm. This may not be the case with smaller protiens? That's something that will need to be tested as well, but I got finals starting on Monday so I need to hit the books this weekend. And figure out what to do about this ticket I got on the way home (94mph in a 60).
Suck magnav0x. And you're the first person to actually beat my speeding record. Mine was 97 in a 65. Anyway, call yourself a decent lawyer. Remeber, anything over 15+ is reckless driving, and that's a felony. Get a lawyer, see if he can't get it reduced. The money you spend on the lawyer will save you far more than paying the inflated insurance rates.
Sorry people but this client is just not going to play nice, I've spent a considerable amount of time since starting the project trying to keep things running. With what is obviously a bad core, bad workunts, supposedly non-existent workunits being released and God (or perhaps Vijay) knows what the hell this lastest "bug" is, I am no longer willing to fight this client to keep producing useful work.
-:Beyond:-
Beyond....how come you get the good bugs!? :shocked:
Sorry to hear you weren't able to get your clients running to par. I finaly got mine the way I want them, took me 3 days into the gauntlet to get em that way though which is quite discouraging. Especialy seeing how it was only 4 machines. F@H has been around for a while and I figured the client would be rock solid and very easy maintanance wise, but I guess all projects have their kinks and wrinkles. I'll stick with it through the gauntlet, but I doubt I'll participate in F@H afterwards.
After running most of the evenimg with no problems get 2 more of the orginal problems cropping up again. F@H "bug" forum
-:Beyond:-
Hey Beyond, on those couple of machines that you are having problems with, you should try to throw them the -verbosity 9 switch, which will display the maximum amount of information while the client is running. Maybe it will give more detail as to exactly what's going on. Just a thought
It is across the whole pharm, every machine has encountered one or more of them problems described in the F@H "Bug" forum thread I started. I (as well as Bruce@FAH Forums) believe it is down to "bad" workunits, and I cannot seem to get to a different server for the "good" workunits. It has been reported that others are experiencing the same type of problems...so it is up to Vijay and company to "correct" things on thier end. Until that happens I am off F@H.
-:Beyond:-
Work units are handed out differently for each server. AKA, each server has it's own type of WUs that it hands out. You may try restricting access to those servers and see if you still have problems...
I need to find which server hands out the big ones
Was over at MacNN.. they have similar problems.
http://forum.folding-community.org/viewtopic.php?t=4380
See Vijay comments at bottom
Agent Smith was right!: "I hate this place. This zoo. This prison. This reality, whatever you want to call it, I can't stand it any longer. It's the smell! If there is such a thing. I feel saturated by it. I can taste your stink and every time I do, I fear that I've somehow been infected by it."
I finally had one of the bugs appear -
the Celeron 708 (Win 98) with the monster 73 pointer slowly died about 72 hours in... Using Taskinfo, i was able to see that FAHCORE_78.EXE had over 10,000 handles open. This made the system very slow and unstable.
After shutting it down politely and restarting, it is back to normal. I think perhaps the faster proteins would finish and restart FAHCORE_xx.EXE before the system ran short of resources.
Posting Permissions
