For caching data, we end up using iRODS.
It's basically the open-source, up-to-date version of SRB (storage resource brokers), very similar to SRM (storage resource management).
iRODs provides more integrated points for data lifetime, access and replication management.
Not the best, but I think it's a bit better than advertising because of the complexity of advertising. Imagine we are giving this to 100 users to run many jobs per day. Because we don't want users have direct access to the machine, or the abililty to write their own condor_submit <submit_file>, therefore, we are moving to iRODS for solution.
I will keep posting when we have our final solution setup. Besides using iRODS, we are also doing a bit of custom caching for other kinds of data.
Thanks
Yeukhon
On Sat, Jan 12, 2013 at 12:21 PM, Dimitri Maziuk
<dmaziuk@xxxxxxxxxxxxx> wrote:
On 1/12/2013 6:26 AM, Matthew Farrellee wrote:
I've had success with startd cron for advertising the contents of a
cache, and highly recommend it over configuration changes.
The script runs as condor (root) user, so security-wise this is worse than letting them 'sudo condor-reconfig' as now they can run anything as condor.
Either way, my point was that doesn't work when you're shipping (flocking, gliding) jobs off-site and have no control over execute nodes whatsoever.
Dima
![[Computer Systems Lab]](http://www.cs.wisc.edu/pics/csl_logo.gif){kind=logo}