It's starting to look like I have my own little forum to ask questions about sge:P
While I'm still processing my backlog of reporting data in arco.... I've been looking at the data that already made it to the DB and it's not turing out to be a godd surprise.
I though I could use Arco to get accounting numbers for our jobs. For example, how many nodes were used by a job useing a memlock and an exclusive complex? Let's say a user request 1024 cores with 6 GB per cores... my compute nodes have 8 cores with 3 GB per core. Arco (as SGE accounting) tells me the job has used 1024 cores that I can multiply per the wallclock to know the usage to report. But it doesn't tell me how many nodes were used: 256 (twice what is getting reported by the number of slots). I know that the sge_job_usage table has one entry for every host.... but that isn't the case anymore since I have enabled 'accounting_summary' in my PEs.
Is there any way, once a job has run to get the hostlist that SGE provided to the job?
If you have your PEs set to accounting summary, then once the job ends there's no longer a record of the specific hosts on which the job ran. If that's important information, you could put together an epilog and squirrels away the pe_hostfile somewhere for later reference, like in another database table.