Page 2 of 3

Re: Segmentation Fault - HELP!

PostPosted: Sat Oct 17, 2009 1:55 am
by Ant P
I actually tried "-smp 4 -verbosity 9" before and got the same crash. Putting -smp at the end doesn't work either.

Re: Segmentation Fault - HELP!

PostPosted: Sat Oct 17, 2009 3:02 am
by bruce
Well, there seems to be a conflict somewhere. Which one of you is correct?

Ant P wrote:I actually tried "-smp 4 -verbosity 9" before and got the same crash.


Kougar wrote:Changing the command to "./fah6 -smp 4 -verbosity 9" fixes the problem and the client now works fine.

Re: Segmentation Fault - HELP!

PostPosted: Sat Oct 17, 2009 11:42 am
by Ant P
Maybe I just hadn't restarted enough times for it to work? :)
toTOW wrote:Try to start the client again. It should start after a few attempts.


Anyway, I can't seem to reproduce it with the A1 core I'm currently running. Works fine.

Re: Segmentation Fault - HELP!

PostPosted: Thu Oct 22, 2009 7:37 am
by Kougar
bruce wrote:
Kougar wrote:Thanks for the reply 7im. I copy 'n pasted the exact command ./fah6 -smp -verbosity 9 out of the Linux SMP install documentation, and did not add any other commands to it. I could test it again if requested, but I'm not sure what would happen with the WU in progress.


Ant P wrote:Never mind... the suggestion to remove the "-verbosity" flag worked for me...


Aha. I'll bet the problem is that the parameter parsing routine is expecting a number after -smp and objecting when it finds -verbosity instead. (This will probably be treated as a low priority bug.)
What happens if you try:
./fah6 -verbosity 9 -smp


I saw your post and changed the command to "./fah6 -verbosity 9 -smp", and it starts without issue. Out of curiosity I went back and tried "./fah6 -smp -verbosity 9" once again and it still resumed folding... maybe it was a first start only issue? It no longer seems to care if I use smp, smp 4, or what order the commands are given. :e?: Yet nothing has changed except a week's worth of folding/reboots.

Re: Segmentation Fault - HELP!

PostPosted: Thu Jan 28, 2010 10:53 am
by Xell
I got this error after upgrading to 6.29. (from 6.02) on a Core2Duo running Ubuntu 9.10. None of the "removing of flags"-trix worked for me and I got the same error no matter how many times i restarted the client. I tried -configonly to run through my configuration, but had to delete the client-cfg file to even start the config, then it booted me with segmentation falt at the end of the config-run.

I tried to revert to static, then alt-version, and suddenly; success!!

Now my question is; is there any way to make 6.29 work or am I stuck with 6.24Beta untill 6.somethinghigerhtan29beta is released?

Re: Segmentation Fault - HELP!

PostPosted: Thu Jan 28, 2010 3:49 pm
by bruce
Xell wrote:Now my question is; is there any way to make 6.29 work or am I stuck with 6.24Beta untill 6.somethinghigerhtan29beta is released?


Maybe we can get Stanford to create 6.29_alt.

Re: Segmentation Fault - HELP!

PostPosted: Thu Jan 28, 2010 4:07 pm
by Xell
It didnt run completely smooth, but at least the _alt the "try-afew-times"-trick worked.

I am considering testing 6.29 on WinXP on the same machine, seeing at they now, supposedly, have the same efficiency. I'll have to see tomorrow. I wagely remember ending my win-smp-stint because of the whole having to have password sett on you user.

Re: Segmentation Fault - HELP!

PostPosted: Thu Jan 28, 2010 6:10 pm
by dschief
If you don't already have it, google Memtest & download & run extended tests. It's possible that you have a memory stik thats border-line

Re: Segmentation Fault - HELP!

PostPosted: Thu Jan 28, 2010 6:43 pm
by djchandler
I got the seg fault (first time ever on Ubuntu) on 9.10 running the 6.24 beta client and while processing an A1 WU. That was the second time the WU had failed, but without the seg fault the first time(AFAIK).

Have since updated to the 6.29 client. No problems so far. YMMV.

For those of you running 9.10, can you tell us which kernel you are using? It may be the kernel that's the problem. Mine is:
2.6.31-17 generic AMD64

And I agree with dschief. It will not hurt a thing to rule out a hardware issue such as bad RAM.

Re: Segmentation Fault - HELP!

PostPosted: Fri Jan 29, 2010 12:13 am
by Xell
I have the 2,6,31-17-generic too, djchandler. This is not a seg fault during a WU run it is right as you start the klient. The log doesn't even get as far as to say Segmentation fault. I had to run it manually to see that segmentation fault was the problem. I comes to the point were it says it is contacting the assignment server and then it shuts down.

This is a run as recorded in the log:

Code: Select all
# Linux SMP Console Edition ###################################################
###############################################################################

                       Folding@Home Client Version 6.29

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: /home/helene/foldingathome/CPU1
Executable: /home/helene/foldingathome/CPU1/fah6
Arguments: -smp 2 -verbosity 9

[07:02:10] - Ask before connecting: No
[07:02:10] - User name: Xell (Team 37651)
[07:02:10] - User ID: 7BACD06412D25A3
[07:02:10] - Machine ID: 1
[07:02:10]
[07:02:10] Loaded queue successfully.
[07:02:10] - Autosending finished units... [January 28 07:02:10 UTC]
[07:02:10] Trying to send all finished work units
[07:02:10] + No unsent completed units remaining.
[07:02:10] - Autosend completed
[07:02:10] - Preparing to get new work unit...
[07:02:10] Cleaning up work directory
[07:02:10] + Attempting to get work packet
[07:02:10] Passkey found
[07:02:10] - Will indicate memory of 2008 MB
[07:02:10] - Connecting to assignment server
[07:02:10] Connecting to http://assign.stanford.edu:8080/



But I will do the memory test. Is it normal that switshing client will trigger a suddenly memory fault?

Re: Segmentation Fault - HELP!

PostPosted: Fri Jan 29, 2010 12:26 am
by dschief
have you always run with -SMP 2 flag? It is my understanding SMP must be a minimum of 4

Re: Segmentation Fault - HELP!

PostPosted: Fri Jan 29, 2010 12:53 am
by bruce
For the A3 core, the default is to use one thread per CPU-core that you have. On a dual machine, that would be 2, and SMP2 does, in fact, run very nicely with 2 threads. In fact, if you specify -smp 4, your performance will suffer.

The only time that you might want to specify a number after the -smp flag is if you don't want to use all of the cpu-cores that are detected, such as if you have HT enabled.

The older SMP clients used FahCore_a1 (Which always uses 4 processes) or FahCore_a2 (which used a minimum of 4 processes).

Re: Segmentation Fault - HELP!

PostPosted: Thu Feb 04, 2010 12:27 pm
by Xell
With regards to -smp 2 i only tried this to see if it would reduce to 2 threads (Core2Duo) as it seemed unnessacery use of ram to start 4 treads on a dualcore. But it did not make any difference. As long as I cannot get the SMP2 client to run the old SMP-client starts 4 threads regardless.

So I guess if I get the 6.29 to work I should remove the 2 after -smp.

Unfortunately, I have not had the time to run a comlete memory test, as my Ubuntu install (9.10) has changed something in Grub so that it cannot find memtest86+ at boot.


edit: found the live-cd and it's running memtest now. But I find it strange that going form 6.02 to 6.24/629 should trigger a memory error. I can still run 6.02 without errors.

Re: Segmentation Fault - HELP!

PostPosted: Thu Feb 04, 2010 4:07 pm
by bruce
On my Core2Duo, when I specify "-smp" it runs very nicely. If I specify "-smp 4" performance suffers.

I didn't check if the memory footprint changed. It might and it might not, depending on how the code is written.

Re: Segmentation Fault - HELP!

PostPosted: Fri Feb 05, 2010 3:42 pm
by Xell
@Bruce: we're talking here about 6.02 or 6.24beta not the 6.29. The 6.29 will not run because of segmentation fault. I think all prior to 6.29 will start 4 threads regardless (?). So when a 6.02 starts 4 threads on a Core2Duo it will take twice the RAM resources the 2 threads.

Anyway; I tested memory. Ran 9 complete runs with memtest86 and no errors, so I guess thats out... luckily for me.