Re: [gembud] Limitations of centrally served gempak?

Kevin,

Testing seems to work, I'm currently running 12 simultaneous sfmap processes with xwindows to my workstation, no errors. I don't mind if 1-2 xw processes fail on a random occasion when doing something like this, it's a built in expectation and usually if that does happen, students can just start over with no worries.

Gotta go run 12 gpend's! ;)

Cheers,
Brendon

Kevin R. Tyle wrote:
Hi Brendon,

I think the # of msg queues by default in CentOS is only 16.

You can up it by editing /etc/sysctl.conf. On our similarly-powered
server, we added these lines to that file:

< # Controls the maximum number of message queues
< kernel.msgmni = 256
< kernel.shmall = 4294967296

Then type "sysctl -p" and that should make 256 message queues available
to you without needing to reboot.

Note though that I've had issues with failed connections to the x server
when multiple clients attempt to access the same x server, causing xw
and gf processes to fail. This happens randomly.

--Kevin

______________________________________________________________________
Kevin Tyle, Systems Administrator **********************
Dept. of Atmospheric & Environmental Sciences ktyle@xxxxxxxxxxxxxxxx
University at Albany, ES-235 518-442-4578 (voice)
1400 Washington Avenue 518-442-5825 (fax)
Albany, NY 12222 **********************
______________________________________________________________________

On Mon, 15 Nov 2010, Brendon Hoch wrote:

Hello,

I'm running Gempak 5.11.4 under 64 bit Centos 5.5 on a server with 16
GB of memory and 8 x Intel Xeon 5450 @ 3.00GHz processors, standard
out of the box kernel, things have been stable for a while.

Historically, I have taught a small class (4-5) where students log in
to a central server via putty and run gempak command line programs
(sfmap,gdplot3, etc.) while piping xwindows to their local screen
(using Xming for Xwindow emulation)

This year, my class size is 11. When I get to about 6 instances of
sfmap running, any additional sfmap attempts gets messages like this:

$ sfmap
Error in message send = 22
itype, ichan, nwords,1,-1,2
Creating process: gplt for queue -1
Error in message send = 22
itype, ichan, nwords,1,-1,3
[GEMPLT -1] NMBRER - Mailbox read.

Interestingly, I do not have this problem with GUI programs like nmap2
or nwx, I have been able to run ~20 simultaneous instances without
incident.

I've looked for renegade gplt processes and ipcs and found nothing out
of the ordinary. System load & memory usage are both reasonable. Is
there a limitation that I am not aware of? Does running in 64 bit have
anything to do with it? Are there kernel options or gempak compilation
options that can be set somewhere to get around this? '

Any advice is greatly appreciated.

Thanks,
Brendon

--
Brendon Hoch
Technology Manager
Judd Gregg Meteorology Institute
MSC 48, Boyd Hall 321A
Plymouth State University
Plymouth, NH 03264
(603)535-2818 Fax: (603)535-2723
http://vortex.plymouth.edu/~bhoch

_______________________________________________
gembud mailing list
gembud@xxxxxxxxxxxxxxxx
For list information or to unsubscribe, visit:
http://www.unidata.ucar.edu/mailing_lists/


--
Brendon Hoch
Technology Manager
Judd Gregg Meteorology Institute
MSC 48, Boyd Hall 321A
Plymouth State University
Plymouth, NH 03264
(603)535-2818 Fax: (603)535-2723
http://vortex.plymouth.edu/~bhoch