Windows SMP Client, general troubleshooting

[Necorum]

Originally Posted by smiffy

Hi Strongwolf
download AMD64 info
http://avala.yubc.net/~lukija/A64Info-beta.exe
and see what the max Async Latency and read preamble are
Slackening is raising the time eg moving MAL from 8ns to 10ns.
Smiffy

thanks for the info smiffy

[Strongwolf]

Thanks that was very helpful. Now to play around with it.

[ThunderRd]

Heh, this is too funny. Both combatants with exactly the same weapon. Smiffy do you have any more secret weapons?

NECO v. SW armed with Max Async Latency and Read Preamble Time

Kinda like Mortal Kombat. Be careful that you don't blow each other up.

[dabaerman]

question for ThunderRd. is SMP used for folding one client with two (both)
cores of a single cpu? will that alone make it fold faster? currently getting
100+ ppd per core @ 50%. nearly the same with my single cores.
I really want to put 3-6 dual core AM2. I have a bunch of DDR and NO DDR2
so 939 will be the least expensive for me to build. if SMP will make a big
difference in folding with dual core, I'm in.

Ron

[Strongwolf]

Originally Posted by ThunderRd

Heh, this is too funny. Both combatants with exactly the same weapon. Smiffy do you have any more secret weapons?

NECO v. SW armed with Max Async Latency and Read Preamble Time

Kinda like Mortal Kombat. Be careful that you don't blow each other up.

Smiffy info/tool was intended for me. Some people just want to win at any price.

[ThunderRd]

Originally Posted by dabaerman

question for ThunderRd. is SMP used for folding one client wit two (both)
cores of a single cpu? will that alone make it fold faster? currently getting
100+ ppd per core @ 50%.

I'm not exactly sure I understand the question. Are you now running the SMP client? If you are there is something dramatically wrong. Of course I don't know what your gear is, but you should be using 100% CPU on a dual core machine, and you should be getting somewhere in the neighborhood of 800 ppd...that's 400 per core. My pentium D's and opterons get between 850-1050 ppd, depending on the WU.

If I understand you correctly, you are running two instances of the normal client...that would yield about what you are saying. Yes, one SMP client for both cores. It was developed for quads, but works fairly well on duals to sync the cores and calculate faster. Go to the SMP and you will increase your production 4-5 times. Do your homework before trying it out, though. I'll help you however I can.

[dabaerman]

yes, running standard FAH client(s). one per core. I will get to know more
before trying it out. the increase in production would be welcomed. I will
hang around the SMP thread for a while.

Thanks
Ron

[Necorum]

Originally Posted by Strongwolf

Smiffy info/tool was intended for me. Some people just want to win at any price.

Muahahaha Releases the hounds !!!!!!


I have an idea!!!:roll ing:

[ThunderRd]

Uh-oh. An idea. That sounds dangerous.

[Strongwolf]

Originally Posted by dabaerman

yes, running standard FAH client(s). one per core. I will get to know more
before trying it out. the increase in production would be welcomed. I will
hang around the SMP thread for a while.

Thanks
Ron

Dude...if I can do it,... anyone can do it.

[Strongwolf]

Originally Posted by Necorum

Muahahaha Releases the hounds !!!!!!


I have an idea!!!:roll ing:

So, there are some neurons alive yet!!!

[Necorum]

Originally Posted by Strongwolf

So, there are some neurons alive yet!!!

Nah those are not needed for this plan

We are the borg

[Strongwolf]

Ran in to troubles again with SMP and Wolf3. I will write it exactly how it appears:

"MPI CH2 Process Manager Argone National Lab removed
MPI CH2 Process Manager Argone National Lab installed but fail to start.
The service did not respond to the start or control request in a timely fashion (error 1053).

account <domain\user> [WOLF3\Juan]"

From that point on I have try various combinations and nada. 3 or four lines explainnig the error that finish something like "the machine actively refuse it" or something.

Installed net 2.0 and have password. I have uninstall/reinstall a couple of times. Where or when I run 'install.bat'. I know is in the DOS install file, but where/when????


Help...help...sos...mayday,mayday,mayday..._ _ _... ..._ _ _...

[ThunderRd]

It would help if I knew how the message came up. Were you starting the client, shutting it down, or was it crunching at the time? What else were you doing at the time? Was this machine already running the client, and did it upload a finished WU already? In other words, was it a properly installed client at some point? Are you overclocking the box? If so, how much? Is the client running as service? A peek at the fahlog in the install directory would also be helpful.

In the absence of that info, I can only recommend to try this:

Close the client, and all instances of mpiexec and fahcore with task manager. Go into the installation directory and find a file called install.bat. Run it. You will see some messages about installing and/or removing MPI. Enter your username [wolf3/juan] and then enter your password two times when prompted.

After that you will see something like "If you see this twice, MPI is working."

If that happens, you should be good to go. Start the client and see what happens. If it doesn't start right away, be patient. Shut down mpi and the cores. Wait a minute, and start the client again. Remember what I told you in an earlier post; sometimes you have to do this a few times when there has been a problem.

If this doesn't work, uninstall the client through add/remove, reboot your machine, delete the installation directory and reinstall. After reinstalling, run the install.bat program again, and try to light up the client. It should start and reach out to an assignment server for the core and a WU.

If none of this works, post again.

Or punt.

[Strongwolf]

Protein: Protein in POPC

Both cores engaged succesfully @ 100%

This is a new build, first time with client/F@H. O/C @2.45

You save my butt again. Thanks!

[Necorum]

i love errors that are not mine

[Daniel ~]

Nice work getting this done guys! ":O}

[Strongwolf]

SMP is not working again. I deal with it later, not felling to well.

[Necorum]

lol :P

[ThunderRd]

First step; REMOVE the overclock and run the client at stock for some time. See if it runs ok then. By some time I don't mean a few hours. I mean a few days to a week. Learn about the client. Observe it. Watch task manager. Find out what it should be doing and what it shouldn't. Get used to it.

I mentioned before that it would be a good idea to let it "settle in" before overclocking.

If you don't want to listen to me, listen to SteveI. He said the same and he's forgotten more than I know about this client.

And if you don't want to listen to Steve, listen to this guy:

Originally Posted by 7im

Networking problems can cause the SMP client to error out. Too much overclocking can do the same thing.

He's one of the site moderators on the folding-community forum. I *think* he prolly knows his stuff.

Relax and go slowly. Don't rush it, it has too many quirks to do that. I waited for a month before I began to overclock. My farm is now overclocked only 10% and I know it should do at least 15-20%, maybe more. But it will stay like this for another month before I begin to push it further.

I deal with at least one or two random problems daily on my boxen. The same kinds of things; EOE's, I/O errors, hung clients, network problems, etc etc. I'm still trying to find out what the "triggers" are for these problems. Some I have learned, some I haven't. But one thing I don't want to do is introduce an entirely new set of problems by overdoing the OC. Better to learn what I can first, then experiment.