CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   CFX (http://www.cfd-online.com/Forums/cfx/)
-   -   CFX problem in LINUX 9 (http://www.cfd-online.com/Forums/cfx/21391-cfx-problem-linux-9-a.html)

Alan June 30, 2005 07:28

CFX problem in LINUX 9
 
Hi,

We've recently setup the LINUX system in one of our new SGI machine (Novell SuSE Enterprise Linux v9, with SGI Propack 4).

When I run a parallel job (transient simulation), the solver will occasionally freeze (when I use "top" to check the status, it shows 0% CPU usage) but there's no error message in Solver Manager. The job just stays on the system for more than 2 days without any sight of progress and I'm forced to kill the jobs without getting any results. I haven't faced any similar problems using a single CPU in this machine. The parallel solver works okay for another SGI machine with UNIX system.

Anyone knows why? Is it a bug? Any solution? Thanks,Alan

Alan June 30, 2005 07:31

Re: CFX problem in LINUX 9
 
The version that I'm talking about is CFX5.7.1.

Alan June 30, 2005 07:41

Missing library in LINUX 9
 
Hi,

I also get some error messages when I start CFX in Linux: Warning: unable to extract version from /lib/libc.so.6.1

Could this be the source of error? How to get rid of this error? Please give some advice as I'm quite new to LINUX anyway. Thanks.

Alan June 30, 2005 07:52

Glenn:please come in..
 
Hi Glenn,

I just found from the old post that you've actually answered a similar question like this. We've plenty of memory (about 60G). My problem only uses up to 1.2m hexa nodes. So it shouldn't be a memory problem. Do you have any idea why the parallel solver is crashed in LINUX 9? Thanks for your help.

http://www.cfd-online.com/Forum/cfx_....cgi?read=8245

Gang June 30, 2005 08:28

Re: CFX problem in LINUX 9
 
Since we are also interesting in SGI machine, your experience will be helpful for us to make decision. What's the model of your new machine, Indigo, Onyx, Origin, or Altix? And what's the CPU in that machine, Xeon, Xeon EM64T, Opteron, or Itanium II? Which message passing package you have used on Linux system, PVM, MPICH, or HP-MPI? Any information will be highly appreciated. Thanks.

Glenn Horrocks July 3, 2005 19:47

Re: Glenn:please come in..
 
Hi,

I've been in the workshop commissioning a machine most of last week so have not been able to get on CFD-Online for a while. I hope you guys did not miss me too much! It's good to get your hands dirty every now and again.

In that old posting I was referring to computer memory, not disk space. If a CFX run does not fit in RAM and it needs to use swap/paging files it will run far slower. If it exceeds RAM memory by a small amount it might work but run a bit slower (don't count on it thought), but if it exceeds memory by a large amount then you have no chance.

LINUX always has its own set of weird and wonderful errors which are incomprehensible. I am not a LINUX guru so I cannot help you with LINUX. We run Windows here as the setup and maintenance required with these systems is far less, at least for small clusters. Also the performance difference between LINUX and Windows these days is small.

Glenn Horrocks

Stevie Wonder July 4, 2005 08:56

Re: Missing library in LINUX 9
 
Hi Alan,


I've posted that issue in the past. As you stated it was on CFX-5.7 version. I remember I was using SuSE Pro 9.0 in that occasion. Now I'm running SuSE 9.3 version and I'm not been running CFX in my linux box, thus I cannot answer you if I still can reproduce this error.

What I can say is at my work we've been running big cases in a Red-hat like linux and this issue doesn't occurs. Another weird errors happen but nothing related to your specific problem. ;-)

About your libs issue what is your output of:
ls -l /lib/libc.so*


Cheers, S. W.

Alan July 5, 2005 09:14

Re: Missing library in LINUX 9
 
Hi,

I'm using CFX5.7.1 anyway. The output is: -rwxr-xr-x 1 root root 2393084 2004-11-06 05:37 /lib/libc.so.6.1

I access the Xterm via putty. Any suggestions? Thanks.

Alan July 5, 2005 09:17

Re: Glenn:please come in..
 
The memory is definitely not a problem for my case. I mean in term of hardware (64GB isn't enough to solve 1.2m hexa nodes? I wonder....) Thanks anyway.

Alan July 5, 2005 09:18

Re: CFX problem in LINUX 9
 
Altix3700 & Itanium II & PVM. Can you see anything wrong here?

Glenn Horrocks July 5, 2005 18:59

Re: Glenn:please come in..
 
Hi Alan,

1.2m hexa nodes would probably run in about 1GB. I assume you have maybe 32 nodes each with 2GB RAM. In that case you should be able to even run this model on one processor, just to check things. Also, I guess the optimum number of processors is likely to only be between 4 and 8 as the model is quite small.

Glenn Horrocks

Gang July 6, 2005 08:23

Re: CFX problem in LINUX 9
 
The installation documents of CFX say that CFX-5.7.1 is known to run on SGI Altix 350/300 with Propack 2 or 3. SuSE Enterprise Linux 9.x & SGI Propack 3 will be certified on Itanium II platform with CFX-10. Therefore, there may be problem to run CFX-5.7.1 on Altix 3700 with SuSE Enterprise Linux 9.x. When you run parallel job, did you run it in local or distributed parallel? Altix architecture should allow you run it in local parallel.

Alan July 6, 2005 10:15

Re: CFX problem in LINUX 9
 
I always run my job with local parallel. I thought Altix 3700 with propack 4 is more stable than Altix 350/300 with Propack 2 or 3...

Alan July 6, 2005 10:18

Re: CFX problem in LINUX 9
 
The job isn't died all the time. Sometimes it will run okay but the other time it may freeze.... Some stability issues here, I think....

santosh August 9, 2005 09:24

Re: CFX problem in LINUX 9
 
We are looking out to buy a cluster to run CFX.Do you suggest an Opteron based cluster or an Itanium.Any specific reasons.Also let me know as to why you chose Altix over other Itanium clusters


All times are GMT -4. The time now is 14:10.