TR Forums

Both of my Q6600 machines have been happily running Notfreds and the A3 SMP2 work units for a few months now. One VM crapped out and it can no longer download v6.29. I think when I originally set them up v6.02 upgraded itself to v6.24, and then to v6.29 once that came out. Anyway one of my machines is still running notfreds virtual folding appliance with SMP2 work units and v6.29 just fine.

I'm having trouble with the Protomol units. After a restart of the PC I'm back to zero. Using the CD and usb stick. Restarts were fine with the other wu's. Checkpoints are set to the standard 15 min.

I just noticed message on the laptop screen that may help (normally screen down). I'll copy it verbatum:

cp: cannot create '/usba/folding/1/work/04/ww.dcd' : No such file or direcory

It goes on to list the same message for the following files:

core.xml, protomol.conf, ww.979.vel, ww.979.pos, checkpt, ww.pf, ww.pdb, scpismQuartic.inp, par_all27_prot_lipid.inp, log.txt

Any way around this problem?

It's a problem with the work units. On every system I've tried, a restart deletes all progress and downloads a new work unit.

farmpuma wrote:

It's a problem with the work units. On every system I've tried, a restart deletes all progress and downloads a new work unit.

Version 24 of the protomol fahcore is *supposed* to fix this problem on regular clients. http://foldingforum.org/viewtopic.php?f=24&t=14567 Delete your fahcore between WUs to prompt the client to download the latest version if not already running v24.

Restarting from 0 on notfred is likely a different issue, and may require a notfred update to handle this new type of fahcore.

Yes, it's finally here:

1 June 10: Upgrade to kernel 2.6.33, Folding 6.29 + build and other minor fixes.

Note it doesn't support -bigadv yet, but at least it is running 6.29 by default on x86_64.

Hooray!

Thanks for that.

A couple of things I'm not sure about:

1) Did something go wrong with the temperature sensor link on the web page the client produces? I haven't gotten it to work on the 2 different machines I'm running the latest (June 1) version on.
2) I can't remember any more, but didn't the ISO generator used to allow for extra parameters to be added? I don't see them now, and I realized today that "-verbosity 9" is missing so I can't track down a different problem which occurred, because I don't know which server I uploaded a WU to, nor the percent of time to deadline remaining. Which means I can't help the Stanford people on foldingforum.org calculate how much bonus was due for a WU that doesn't appear to have gotten one.

I am more and more confident that something is wrong with the temperature sensor code. On the 1 machine I have presently running notfred's ISO, every time I click on the temp sensor link, the browser acts as though I'm loading a stupendously large web page or something from a slow server, anyway, and then the machine's folding speed seems to reduce by about 95%.

Clicking on the reboot link eventually solves the slow speed problem, at least.

As for the parameter thing, I guess the solution for the present is to manually login to the machine, killall the client after it hits a checkpoint, and run -configonly and put the extra parms in there, and then restart the client. The problem with that is you have to do it every time the machine restarts.

I've had another report (by PM) of issues in the temp sensor webpage causing instances of httpd to sit there hogging CPU. If you have the console enabled, you can find the PID of the hogging ones with top and kill them manually by hand. I need to get some time to debug what is going on in the temp sensor code. BTW what CPU are you seeing this issue on?

The only one I'm running your ISO on presently is an Intel Q8400 (quad core, stock speed is 2133 MHz but it's reliably running at 3080 MHz for the time being). I had been using it on an AMD X2-3200 overclocked a few dozen MHz, but shut it down because I need the power for the absurd i7 power hog I'm in the process of tearing my hair out over. It runs hot, and uses power like a 12 core Xeon based on a P4 or something. But that's for another thread. Back to the AMD: I'm pretty sure it gave me the same problems the Intel is giving on the temp. sensor link, though I don't remember the change in performance with it, just the fact that the page didn't load. It's odd that clicking the back button on one's browser wouldn't eventually eliminate the bad process, or at least make it use less processor time.

It appears to have been a bug in Busybox, I've upgraded from 1.16.0 to 1.16.2 and it now seems to be working so I've released a new version of my stuff.

Hi, notfred!
I'm testing the diskless for a few days now and i run it natively on my desktop pc in SMP mode. What I observed nearly from the start is that temperatures reported for the 2 cores of the cpu are about 10*C higher than real temps. I've run for a while the smp win client and my temps won't go higher than 70-74*C. The values indicated when running the smp in the diskless are about 80-84*C, which is obviously out of reality. In this setup (oc @3200MHz) the highest cpu temp it have reached is about 79*C while running Intel Burn Test v2.
Is there a way to get real temps of cpu being reported by the diskless app?
oh, and one more question: is there a posibility to add another parameters like "ignore deadline", because i would like to run performance tests on a captured workunit. Regarding this, It would be useful (as someone else already requested) to not delete all the files needed at runtime, every time it quits.

my system: E2180 cpu and GA-P31-DS3L mainboard, with 2x2GB DDR2 1066.

thanks in advance,
Edy

I believe the temperature values are correct, they are coming from the temperature sensor built in to the processor die. Linux tends to make harder use of the CPU than Windows, particularly when things are as barebones as they are on this distribution. I expect it to definitely be higher than Windows folding, and as the Intel Burn Test seems to just be doing Linpack, I'm not surprised that Linux folding is higher than that. The webpage is just running the "sensors" command, if you run it from the command line you will see the same temps.

If you want to ignore deadline, after it is setup and running, with the command line you can go in and stop the running folding process, put a new client.cfg on there that contains and manually start the folding.

The check_hang.sh process does the cleanup of the work directory. Remember that everything is on a RAM file system so left over files take away memory available to fold with, that's why it cleans up aggressively as otherwise you can end up running out of memory. If you want, you could always use the command line to kill the check_hang.sh and then it will not cleanup completed work units.

Thanks for the quick reply
I forgot to mention that early yesterday I enabled the CPU temperature alarm in bios and set it to 80*C. I tested it another time and it worked, at least for the 70*C threshold.
So alarm should go off for 80*C+ either, in case the temperature reach that level.
It didn't happen in over 5h of testing. During this time it showed temps as high as 84*C, but no alarm .

The BIOS temperature may be coming from an external sensor, I'm not up on how the BIOS reads the temperature, especially when the processor is up and running in an OS. The Linux kernel is reading the temp in the same way as the Windows Core Temp program, they explain more at their webpage http://www.alcpu.com/CoreTemp/

I just tried Core Temp. It's showing the same temperature like Everest or CPUID HWmonitor do. So i think there could be something wrong when using diskless.
Tomorrow I'll try the linux smp client on an Ubuntu setup and see what i get. I'll Keep you posted!
Thanks again.

LE: Today is a hot day and i managed to reach and go beyond 80*C in windows doing some stress testing, and the alarm went off immediately!
Tested again folding smp on diskless and no alarm, although it showed me 84-87*C ! There must be something wrong with these values.

Not insisting on temps anymore, now that i know in linux is 8 degrees higher than normal (bios and windows) and it's almost sure that bad values are those in linux (any, not only diskless version).
But I have another question regarding folding on diskless: how to make fah run automatically on screen, not in background ?
I tested it and runs ok if started manually, but i would like to have it starting up and showing progress on screen also, not only on the web server.

Hi NotFred, Will your folding disk allow 4 instances of GPU folding? How much RAM would be required? I am shipping a folding node to Emcubed that will have two 9800GX2s and he wants to use your folding disk.

Thank you for your support.

Sorry but no GPU folding support in it, just SMP.

Rats! Will it support 4 SMP instances?

notfred wrote:

Sorry but no GPU folding support in it, just SMP.

The Windows SMP discussion has been split out to this new thread.

I've been running your ISO as the Linux system on a VMware Player setup since Thursday night, and have a few things to ask:

1) How does one get the time to show as UTC instead of local time?
2) The backup.sh script seems to require "bootloader type" to = 49, while mine is giving 51 as the result. I don't know what this number means, but it's causing the error message "Hard drive image is corrupted, not backing up" to post every time a backup is supposed to be done. But it appears like the backup.tar file is being generated, so can I just change the number to 51? It's late so I've forgotten and haven't tried to figure out what it's failing to do again yet, but... (I forget what else I was going to say).
3) There seems to be a problem finding the VM's "machine" on my network. The host machine can find it, but other boxes can't. What am I doing wrong?

I've changed the thing around to download the 6.34 version of the Linux client, and it looks like I could probably change the CD generator code to make most of the changes I would like, including allowing the user to add extra parameters to the call to fah6 as -configonly lets one do. That would help people like me make it run the -bigadv stuff, which would help those using your ISO significantly. Do you have time to update your web page for this stuff if I submit the changes to you?

Thanks.

I haven't done anything on this in a very long time. If you are interested in helping develop it I can try and get some time to work on it again, although I'm visiting family overseas currently so it will be a couple of weeks.

1) The time will be UTC if you run on a bare machine. I suspect you are running on a Windows host and Windows runs local time on the hardware clock so VMware is probably passing that through.

2) I'll check this when I get back home, I think it might be doing the check based on a USB stick bootloader value.

3) I suspect you have the VM set to NAT'd networking and not Bridged networking. In NAT'd networking there is a router between the VM and the rest of your network, in Bridged the VM is directly on the wire.

With respect to #3, it was at the default NAT. Changing it while it's running seems to reset things, but not allow me to connect to the outside world using the "wget" scriptlet I have to download the Linux 6.34 client as a test. I'm not sure if restarting the entire VM would make it work or not, but I'm not ready to test that until the current WU finishes in ~14 hours.

I'll play with the CD page at some point, though I've managed to avoid that sort of thing so far in my life.

Thanks for the reply.

How are you guys booting these machines with no video card installed? I thought the machine would halt when it detected there was no card. Do you need to choose a particular motherboard or BIOS option to get this to work?

I can see the advantage of one less expensive part and the reduction in power. Even onboad graphics much chew up some decent amount of power when running 24/7 x 365.

Dirge wrote:

How are you guys booting these machines with no video card installed? I thought the machine would halt when it detected there was no card. Do you need to choose a particular motherboard or BIOS option to get this to work?

I don't use Notfred's image, but when I need a "headless" system I generally just choose a motherboard with onboard graphics. Yes, you are correct that most non-server boards will get upset if they don't see a video card...

Dirge wrote:

I can see the advantage of one less expensive part and the reduction in power. Even onboad graphics much chew up some decent amount of power when running 24/7 x 365.

But small relative to the amount of power being chewed up by a CPU running at 100% load.

No more diskless folding farms? I just got some CPU's hooked up and was wondering what to do with them.