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

Re: [Condor-users] Site-wide want_remote_io=false possible?

Hi Todd,

Todd Tannenbaum wrote:
> Well, you can tell condor_submit to add this setting by default into 
> each submit file.  If you are able to edit the condor_config file(s) of 
> all machines where your users run condor_submit, the following entry 
> should do the job:
>     SUBMIT_EXPRS = WantRemoteIO=True
> (the Manual has more details on SUBMIT_EXPRS)

I guess I was searching (again) for the wrong expressions. Thanks for
the pointer!

> No, they can write to local disk, but realize in standard universe if 
> they write to local disk then Condor will not do any work to send the 
> files back to the submit machine; so writing to local disk is fine for 
> temporary files (who will clean them up?) but not such a good idea for 
> output files.

Good, right now that's not much of an issue, but we will start a cleaner
from time to time to get rid of old files.
> Maybe your UID_DOMAIN and FILESYSTEM_DOMAIN settings are incorrect?

No both contain to the same string (tested on submit hosts and a few
compute nodes).
> Basically, in standard universe if the UID_DOMAIN and the 
> FILESYSTEM_DOMAIN of between the submit and execute machine are both 
> equal, then Condor should use the shared filesystem to access the file 
> and not use remote I/O.  Does your submit machine and execute machine 
> mount the same shared filesystems on the same mount points?  If yes, 
> then I would make sure the shared filesystem settings are correct and 
> then Condor should "do the right thing(tm)" and not require you to 
> disable remote I/O all the time.

Well, the user has her home file system under /home/<username>
throughout the cluster, but since this is on NFS, condor log files are
usually stored on /local on the submit machine which is not shared under
this name throughout the cluster. Are the log files an exception or do
these have to have the same path everywhere?

Thanks already!