-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 05/12/07 02:51, william pursell wrote: > Ron Johnson wrote: > >> On 05/11/07 19:36, s. keeling wrote: >>> Ron Johnson <[EMAIL PROTECTED]>: >>>> On 05/11/07 12:49, s. keeling wrote: >>>>> Ron Johnson <[EMAIL PROTECTED]>: > >>>> How do you limit the number of batch jobs that can run at any one >>>> time? (If the "job limit" of a queue is 4 and you submit 20 jobs, >>>> the first 4 jobs grab the execute slots, and the remaining 16 wait >>>> until execute slots open up.) >>> Meaning, you'd prefer that all 16 jobs run concurrently? That sounds >>> sub-optimal (for most eventual users, at least). >> >> No. Exactly the opposite. >> >> Think of a bank with 4 tellers. If 20 people walk into the bank, >> the first 4 get to the tellers, and the other 16 wait in the queue. >> As a customer at a teller finishes his business at a teller and >> walks away, the next customer in line goes up to that teller. > > I'm probably immersed too much in *nix, but I'm thinking, "Why?" > If you have 20 jobs and 4 cpus, you schedule them all to run > concurrently and you let the scheduler worry about the details > of what runs. If you really want to ensure that you don't > take a hit for context switches, then you run them with a > high absolute priority, via sched_setscheduler(). Most batch > processing is I/O bound anyway, so this won't really have > any impact.
That's the theory, anyway... > But, basically, it sounds like you want the > batch manager to take over the job of the scheduler. I > don't see the advantage of that. The "4 slot limit" was just an example. Right now, we've got dozens of jobs running simultaneously. - -- Ron Johnson, Jr. Jefferson LA USA Give a man a fish, and he eats for a day. Hit him with a fish, and he goes away for good! -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFGRZE0S9HxQb37XmcRAvYaAJ97YV8/NSkb1MbMAaRVTpl1I4xR0ACg0KTJ hA0gev1ntMv+QjraosS0YtE= =wEsV -----END PGP SIGNATURE----- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
