[lug] Redhat Ent5 madness again

[Hugh Brown]

strace -fvvvvto /tmp/foo.strace -s 2048 fooscript

The file /tmp/foo.strace will have the output with timestamps for each system call made.

Hugh

On Thu, Mar 20, 2008 at 02:44:50PM -0600, Steve A Hart wrote:
> Looks like I jumped the gun saying the problems were solved.
>
> They appeared to be solved but have resurfaced and I'm going to start 
> loosing hair out of shear frustration on this one.
>
> Quick overview again.....
>
> Server is running RHEnt4
> Clients are running RHEnt5
>
> all programs (firefox, thunderbird, matlab. IDL, etc) are compiled for 
> RHEnt4 and are in /usr/local on the server which is then automounted on the 
> clients.
>
> Now the problem:
> To rule out any type of gnome or KDE issue, I did a ctrl-alt-f6 and logged 
> in that way.  So no graphical environment whatsoever.
> 1. start matlab with "matlab -nojvm" from within /tmp (so that everything 
> is local to the system itself except matlab which is automounted).
> 2. ran the following matlab script
> -------
> tic
> [status,result] = system('ls')
> toc
> -------
> On Hugh's setup (thanks again for testing this!) he got a time of 0.3second 
> which is what I was seeing before I upgraded the clients to RHEnt5 from 
> Fedora Core 6.  On my current clients, it's varying from 2-11 seconds.
>
> If I reboot my client and try it, it will work in 0.3 second but an hour or 
> so later it will go back up to 2-3 seconds or longer.
>
> It's not an nfs issue between the client and server (i.e. automount 
> related).  We monitored the network activity very closely and found that at 
> no time did it show any problem.
>
> It appears to be an issue with any type of program calling a Linux system 
> command.  I say this because another user is running a shell script that 
> calls /bin/sed as well as /usr/bin/g++ which is the C++ compiler.  This 
> script (that on FC6 ran smoothly) is now running exceptionally slow on 
> Ent5.  No matter what I do (i.e. reboot), this script runs like a snail.
>
> I'm quickly running out of ideas on this and could use some more brain 
> power to put this one to bed for good.
>
> Any help/ideas would be greatly appreciated!
>
> Steve
>
> -- 
> Steve Hart
> Systems Administrator
> Colorado Center for Astrodynamics Research
> University of Colorado Boulder
> shart at colorado.edu
> (303)492-8109
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug
> Join us on IRC: lug.boulder.co.us port=6667 channel=#colug
>