[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [condor-users] Condor installation



On Fri, Dec 26, 2003 at 09:06:36AM -1000, ron.viloria@xxxxxxxxxxxxx wrote:
> Hi,
> 
>     In the past we were trying to run linux based compute nodes
> simultaneously with windows based compute nodes such that jobs would load
> balance across all of them. We are using a Windows based central manager.
> Interesting situations arise as in the Windows setup the file transfer
> mechanism is used and in the linux setup a shared filesystem is used. Is
> this now possible or can someone direct me to a link or documentation that
> can help with this configuration recipe? Can someone help identify issues
> with this?
> 
>     To clarify, we want to submit a job from Windows Server (the submitter
> and central manager are the same machine, as it is also the central
> webserver for a web submission system), but we want the job to be
> distributed to a heterogenous set of compute nodes which currently are
> Windows and Linux.
> 

When you submit jobs, you can specificy that some details of the job need to
be left blank until they're actually matched. This is done with the 
$$() operator in the submit file, which means "take the attribute from
the machine ad and use it." For example, imagine you have a simulator
that can run on Linux or Windows. If you have them both named simulator.LINUX
and simulator.WINNT50, you could do something like:

universe = vanilla
copy_to_spool = false
executable = C:\tmp\simulator.$$(OPSYS)
requirements = (Memory > 512) && ( (OpSys=="LINUX") || (OpSys=="WINNT50") )
log = simulator.log
output = simulator.output
queue

Both Linux and Windows support transfering of files - you could add
something like:

transfer_input_files = input.dat
transfer_output_files = results.dat
should_transfer_files = yes
when_to_transfer_output = on_exit

means that every time the job runs, Condor will copy input.dat to the 
execute machine, and copy results.dat back. The $$() stuff will work here, so
if you want, you can have different inputs for different machines:

transfer_input_files = input.$$(OPSYS)

If there is a shared filesystem, Condor can detect this at match time and not
transfer the files - see the "IS_NEEDED" option to should_transfer_files
submit command. Because you're submitting from Windows, getting a shared
filesystem working between both Linux and Windows is challanging (but 
probably doable), and if you can use file transfer you should.

The Condor file transfer mechanisms are documented here:
http://www.cs.wisc.edu/condor/manual/v6.6/2_5Submitting_Job.html#SECTION00354000000000000000

-Erik

Condor Support Information:
http://www.cs.wisc.edu/condor/condor-support/
To Unsubscribe, send mail to majordomo@xxxxxxxxxxx with
unsubscribe condor-users <your_email_address>