At RAL we’re migrating from Torque/Maui to HTCondor – currently we’re at 50% Torque and 50% Condor, and will be 100% Condor by early November. We are using 3 ARC CEs and 2 CREAM CEs.
It’s actually not that hard to get an EMI-3 CREAM CE to work with HTCondor. BLAH still actually supports HTCondor, so the job submission side of things works immediately. The problem is that Condor was removed from some of the YAIM functions, the script which is used to publish dynamic information no longer exists, and there is no APEL parser for Condor. We started with the dynamic info publisher script from a much older CREAM CE release (which still supported Condor), and modified it so that it works with an EMI-3 CREAM CE. We also wrote a simple script to parse the Condor history files and re-write pbs style accounting files, so that the standard pbs EMI-3 APEL parser can be used.
Apart from us, I think Milan is the only other site using CREAM CEs and Condor.
With the Condor batch system at RAL, ATLAS and CMS are only using the ARC CEs. LHCb have recently started to use them, however ALICE are unable to and don’t have the effort to get them working, so they are using the CREAM CEs. Note that jobs can be submitted to ARC CEs via the WMS in the same way as with CREAM CEs (provided the WMS is using a recent version of Condor internally).
We definitely prefer ARC CEs to CREAM – they are much simpler, have just a single config file, and don’t require YAIM at all. Also, they can submit accounting records directly to the central APEL service, so you don’t need to have an APEL publisher node at all. The only downside I suppose is that the Condor backend scripts for ARC were written many years ago, and we had to make some minor tweaks. I’ve submitted tickets for all the issues, so eventually a future release will work correctly out-of-the-box. 2 other UK sites are in the process of moving from CREAM + Torque to ARC + Condor.
Let's suppose there's an existing system with LSF as batch system and layers like CREAM as computing element, and ICE for the WMS.
If I'd like to replace LSF with HTCondor, what can I use instead of CREAM in order to stay grid-compliant? (as far as I understood, CREAM doesn't support HTCondor as a batch system)