Hi Jenny:

You should be able to run both SGE and Torque at the same time (though I
haven't tried it) since they are drastically different systems.  The
only thing to keep in mind is that they use similar binary names for
commonly used programs like qsub and qstat, so make sure you know which
one you're executing.

But it might be just as easy (and cleaner) to install Torque and SGE on
different installations (machines), and then you can see their
performance side by side.

Cheers,

Bernard

> -----Original Message-----
> From: Jenny Aquilino [mailto:[EMAIL PROTECTED] 
> Sent: Friday, September 03, 2004 12:22
> To: Bernard Li; [EMAIL PROTECTED]
> Subject: RE: [Oscar-users] How do you make master node a 
> compute node too?
> 
> Oh, that's great news.  I was going to check to see if the 
> opd would list Torque or SGE but you answered that for me.  
> Before I sent off my last e-mail saying that configuring the 
> master as a client through the install_server GUI killed my 
> setup, I tried it again since I've learned a lot since I last 
> tried that and sure enough it did the same thing again.  Doh! 
>  So I'm currently in the process of rebuilding that machine 
> again but I'll try installing Torque after I've got it up 
> again and see how that works.  I just saw Arnie's post and he 
> seems to be happy with SGE so I can try that too.  I hope 
> there are no issues with having several schedulers installed. 
>  I would turn the daemons off for the others but still.  Ok, 
> well it looks like my system just finished kickstarting so 
> I'm gonna try this again. 
> Thanks for all the info.
> 
> -Jenny  =)
> At 12:06 PM -0700 9/3/04, Bernard Li wrote:
> >Hi Jenny:
> >
> >I believe PBS Pro is free for educational uses, so perhaps 
> you guys can
> >get it for free.
> >
> >Torque is basically OpenPBS but with it all patched up (and updated
> >quite regularly too).
> >
> >Sun Grid Engine is getting more and more popular (and my personal
> >preference).
> >
> >You can actually d/l and install Torque from OPD (OSCAR Package
> >Downloader).  I am working on creating a package for SGE as well but
> >that has been pre-empted by other projects I am working on.
> >
> >Cheers,
> >
> >Bernard
> >
> >>  -----Original Message-----
> >>  From: Jenny Aquilino [mailto:[EMAIL PROTECTED]
> >>  Sent: Friday, September 03, 2004 12:03
> >>  To: Bernard Li; Michael Edwards; [EMAIL PROTECTED]
> >>  Subject: RE: [Oscar-users] How do you make master node a
> >>  compute node too?
> >>
> >>  Hi Bernard,
> >>
> >>  That is good input.  I went to their site to try and get some
> >>  documentation and it looked like OpenPBS is no longer so
> >>  Open.  =( Professional PBS appears to be what they're
> >>  throwing their effort behind now.  I'll check out the other
> >>  batch schedulers you suggested.
> >>  I can still use my same infrastructure that was setup with
> >>  OSCAR and just use cexec or something to push out the new
> >>  scheduler to the other nodes...once I have them.  Thanks for
> >>  the feedback.  =)
> >>
> >>  -Jenny  =)
> >>  At 11:28 AM -0700 9/3/04, Bernard Li wrote:
> >>  >Hi Jenny:
> >>  >
> >>  >Just throwing in my 2 cents here.
> >>  >
> >>  >If you are new to batch scheduling systems and you don't have a
> >>  >preference, I would recommend using something like Torque or SGE
> >>  >(instead of OpenPBS).  OpenPBS is no longer being maintained
> >>  and there
> >>  >are much better choices out there.
> >>  >
> >>  >Cheers,
> >>  >
> >>  >Bernard
> >>  >
> >>  >>  -----Original Message-----
> >>  >>  From: [EMAIL PROTECTED]
> >>  >>  [mailto:[EMAIL PROTECTED] On
> >>  Behalf Of  Jenny
> >>  >> Aquilino
> >>  >>  Sent: Friday, September 03, 2004 11:08
> >>  >>  To: Michael Edwards; [EMAIL PROTECTED]
> >>  >>  Subject: Re: [Oscar-users] How do you make master 
> node a  compute
> >>  >> node too?
> >>  >>
> >>  >>  Hi Michael,
> >>  >>
> >>  >>  Wow, thanks for the quick response.  =)  Yeah, please
> >>  don't  break
> >>  >> your production cluster just to check this out.  I will 
> >>  look at the
> >>  >> installer again to see if I see that check box  you're
> >>  talking about.
> >>  >>  It actually sounds really familiar.  I installed an OSCAR
> >>   cluster
> >>  >> about 3 years ago and I think I remember seeing that  
> then but I
> >>  >> can't seem to find it now.  I did try defining the  
> master nodes
> >>  >> private interface as node0.cluster using the 
> >>  install_server GUI with
> >>  >> horrific results.  Yeah, I won't try  that again.  It 
> completely
> >>  >> broke my access to the oda  database so I couldn't 
> back out of the
> >>  >> installation or fix  it.  I spent quite a bit of time
> >>  trying differen
> >>  >> things like  granting access to the [EMAIL PROTECTED]
> >>  user but
> >>  >> it  still wouldn't work so I finally just threw my 
> hands up and 
> >>  >> rebuilt the system.  =P
> >>  >>
> >>  >>  As for just geting mpich to work on its own, it appears
> >  > that  it is
> >>  >> configured correctly after the OSCAR installation so  
> that isn't
> >>  >> really a huge problem.  Well, except for the fact  
> that it is only
> >>  >> seeing one processor for use.  If you have  any idea 
> why that is
> >>  >> happening please let me know.  I would  really like to
> >>  have all the
> >>  >> management stuff working though  which is why I'm 
> still pushing to
> >>  >> try and get the xpbs stuff  setup correctly.  If you have
> >>  any other
> >>  >> ideas for me, that  would be great.  Thanks Michael.
> >>  >>
> >>  >>  -Jenny  =)
> >>  >>  At 12:33 PM -0400 9/3/04, Michael Edwards wrote:
> >  > >>  >I am fairly sure there is a radio box somewhere in 
> the  install
> >>  >> process  >that says "Use head node to compute".  I will
> >>  try and see
> >>  >> if  I can find  >it if I can find a computer I don't mind
> >>  breaking... 
> >>  >> Don't want to  >mess with my production cluster, at least
> >>  one of the
> >>  >> installer steps  >does things even if you hit cancel :)  >
> >>   >If you
> >>  >> just need to use lam (or I assume mpich, never tried  
> installing 
> >>  >> >that one), it is quite easy to install stand alone and 
> >>  configure to
> >>  >> run  >on one computer.  OSCAR is mainly handy for 
> installing the
> >>  >> scheduler  >and resource managers.  Then you could go back
> >>  and start
> >>  >> over with  >OSCAR once you get some compute nodes.  
> Just a thought
> >>  >> for a  quick fix.
> >>  >>  >
> >>  >>  >Unless there is something else in OSCAR you would need for 
> >>  >> development,  >I can't think of anything off the top of my head
> >>  >> though.
> >>  >>  >I could walk you through using lam on just one node if
> >>  you  need a
> >>  >> hand,  >I have done it before while testing things.
> >>  >>  >
> >>  >>  >On Fri, 3 Sep 2004 09:05:44 -0700, Jenny Aquilino 
> >>  >> <[EMAIL PROTECTED]> wrote:
> >>  >>  >>  Hi,
> >>  >>  >>
> >>  >>  >>  I was hoping that someone out there might be able to
> >>  help me out 
> >>  >> >> with  this question.  I am currently working on a 
> rather strange
> >>  >  > >> cluster...it's only one node.  I know that sounds strange
> >>  >>  but there
> >>  >>  >> is a reason for it.  The user only has one system to
> >>  start  out
> >>  >> with  >> but would like to add more nodes very soon.  The
> >>  problem  is
> >>  >> that he  >> can't wait for the other nodes to come on-line
> >>  before he
> >>  >> starts  >> developing on it so I need to make this one
> >>  node cluster 
> >>  >> functional  >> for him.
> >>  >>  >>
> >>  >>  >>  I saw the FAQ on the OSCAR sourceforge website 
> that says  to
> >>  >> modify  >> the /var/spool/pbs/server_priv/nodes file and
> >>  then restart
> >>  >> the  >> pbs_mom, pbs_server in order to make the master a
> >>  node so  I
> >>  >> did that.
> >>  >>  >>  When I run "pbsnodes -a" it does reflect my one node.
> >>  >>  However, when
> >>  >>  >> I run xpbs to see how many processors are allocated to my 
> >>  >> server, it  >> reflects 0.  It also shows the workq as 
> not having
> >>  >> access to any  >> processors.  I added a print 
> statement to one of
> >>  >> the mpich  example  >> programs to print out the number of cpus
> >>  >> recognized by the  >> "MPI_Comm_size" command and it
> >>  returned 1 when
> >>  >> in fact my  system has  >>  2 processors.  Does anyone
> >>  know why this
> >>  >> might be?  Is  one of the  >> processors on the master 
> node always
> >>  >> reserved?  More  generally, is  >> there a set of steps
> >>  anyone could
> >>  >> recommend to be able to get the  >> master node setup like
> >>  a client
> >>  >> as far as the monitoring and  >> scheduling tools go?  I
> >>  did go into
> >>  >> qmgr and by hand can  enter the  >> information but it 
> seems that
> >>  >> when other nodes get built, that  >  > information is
> >>  automatically
> >>  >> populated into the tables  and I would  >>  really like it
> >>  if I could
> >>  >> get the master to populate this  >> information  on itself
> >>  when it is
> >>  >> setup to also be a compute node.
> >>  >>  >>
> >>  >>  >>  I hope this e-mail makes sense.  I'm a little crazy after 
> >>  >> several  >> nights of tossing and turning trying to figure
> >>  out why 
> >>  >> things aren't  >> working the way I would expect them to. 
> >>  Thanks in
> >>  >> advance  for any  >> help you can offer.  =)  >>  >> 
> >>  -Jenny Aquilino 
> >>  >> =)  >>  >>  
> -------------------------------------------------------
> >  > >>  >>  This SF.Net email is sponsored by BEA Weblogic 
> Workshop  FREE
> >>  >> Java  >> Enterprise J2EE developer tools!
> >>  >>  >>  Get your free copy of BEA WebLogic Workshop 8.1 today.
> >>  >>  >>  http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click
> >>  >>  >>  _______________________________________________
> >>  >>  >>  Oscar-users mailing list
> >>  >>  >>  [EMAIL PROTECTED]  >> 
> >>  >> https://lists.sourceforge.net/lists/listinfo/oscar-users
> >>  >>  >>
> >>  >>
> >>  >>
> >>  >>
> >>  >>  -------------------------------------------------------
> >  > >>  This SF.Net email is sponsored by BEA Weblogic Workshop
> >>  FREE  Java
> >>  >> Enterprise J2EE developer tools!
> >>  >>  Get your free copy of BEA WebLogic Workshop 8.1 today.
> >>  >>  http://ads.osdn.com/?ad_id=5047&alloc_id=10808&op=click
> >>  >>  _______________________________________________
> >>  >>  Oscar-users mailing list
> >>  >>  [EMAIL PROTECTED]
> >>  >>  https://lists.sourceforge.net/lists/listinfo/oscar-users
> >>  >>
> >>  >>
> >>
> >>
> >>
> 
> 
> 


-------------------------------------------------------
This SF.Net email is sponsored by BEA Weblogic Workshop
FREE Java Enterprise J2EE developer tools!
Get your free copy of BEA WebLogic Workshop 8.1 today.
http://ads.osdn.com/?ad_idP47&alloc_id808&op=click
_______________________________________________
Oscar-users mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/oscar-users

Reply via email to