On 03/25/2013 09:37 AM, Rich Pieri wrote:
--On March 25, 2013 9:10:57 AM -0500 Christopher Harrison <harrison@xxxxxxxxxxxxxxxx> wrote:I might recommend a SSD for the submit machine's "spool" area. I have experienced extremely fast io access, reduced latency and improvedthroughput just by installed the ssd's. Just one item you might want toconsider.
Note that, depending on your job mix, the HTCondor submit "spool" area can get kind of big, maybe bigger than your SSD. If you can't fit the whole spool on SSD, much of the benefit can still be gained by just putting the job_queue.log file there. This is the file that the scheduler uses as the persistent store, and is fsync'ed very frequently, so it is very latency-sensitive. JOB_QUEUE_LOG is the name of the config parameter.
-greg