[GE users] Other documentation for Grid Engine 6 ques setup

Marconnet, James E Mr /Computer Sciences Corporation james.marconnet at smdc.army.mil
Fri Feb 4 15:55:36 GMT 2005


 
Reuti's suggestion (1) is interesting, and I have an idea what he's talking
about, but I think it would be tough for us to adminster such ques due to
frequent changes. Easier for us to just let the users write their scripts to
submit their jobs. Some of them edit their qsub .sh files manually. Some use
MATLAB to generate their multiple qsub .sh files and also a script to qsub
all of them. Others write qsub scripts that write multiple qsub scripts on
the fly based on the assigned job number. Everyone has their modus operandi
set up and working, so I don't want to suggest changing anything beyond the
que of the day/week/whatever.

The explanation (2) how to find/access things like: "sched-conf(5)" led me
to typing in "man sched-conf" where I had been typing in the full "man
sched-conf(5)". Very confusing why they put the (5) on the end when man
cannot find anything that way. And I did not see any (5) within the manpage.

In the sun GE User's and Adminstrator's documentation, I found a total of 2
lines about nice. When I tried "man nice" there was a little more, but
that's still fuzzy to me. Using QMON, I can set nice for a specific que
and/or a specific node within that que between -20 and 20 by clicking the up
and down-arrows, but some numbers sometimes don't display, instead giving an
error message in the window from which I ran qmon, something about "no font
existing for that particlar character". Very confusing. Anyway -20 to 20
nice range in qmon is different from what they say in "man nice". Found a
mention, here I think, about -nice being not allowed by SGE as a "safety
factor". Confusing to me. For now I'm using 0 to 20 nice to be safe. I would
have appreciated a paragraph on nice and how it works in the sun
documentation.

Load thresholds and such are becoming more clear to me. Now if I can figure
out if the right number to use is 1 or 4 because we have some
dual-processor, hyperthreaded nodes in the cluster set to 4 units.

The link (3) to IBM tutorials is nice. I read thru them all and had no idea
which tutorial would be helpful to me, specifically for nice. I bookmarked
it for later study.

Thanks!
Jim Marconnet

-----Original Message-----
From: Reuti [mailto:reuti at staff.uni-marburg.de] 

<snip>

> We users are all friends, and we want to keep it that way. But we all 
> have our own work to get done. We want to keep it somewhat simple 
> using several different (primary and secondary) ques that the users 
> would routinely submit to, rather than having a lot of job-specific 
> user input required when submitting each individual job. Ideally just 
> pick your que based on who you are and whether your job is long- or 
> short-running. And perhaps have a lower-priority que for the jobs you 
> want to run, but when you are not up against a deadline.

(1) this is an interesting approach. If you have only a really few fixed
applications, it may be a solution to put the job script (which is fixed and
setup by the administrator) in a startup method of a queue. You have a job
of type x - submit to queue x. And the mentioned script in the qsub command
is just the input file to the program. But this is an advanced setup.
 
<snip>

> I am especially struggling with understanding and the interations 
> between "nice", "priorities", "load_thresholds", "subordinate list", 
> and "over-subscribing". These are all in the Sun documentation and 
> have been discussed here at length, but so far I've not been able to 
> put it all together into something coherent.
> 
> And I keep seeing references to things like: sched-conf(5) manpage, 
> but have no idea what they are talking about or where to find this, 
> much less how to set it.

(2) It's a reference to the man pages, which are most likely already
installed, when you have SGE already running. Try: man sched_conf  But to
"nice" and "load_thresholds", you'll need some insight to the used operating
system. E.g. 
(3) for Linux you may look at IBM Developer documentation 101: 
http://www-106.ibm.com/developerworks/views/opensource/tutorials.jsp

Cheers - Reuti

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list