Hi, Am 03.07.2014 um 16:46 schrieb Dan Hyatt:
> I have been trying to configure SoGe with qmon, with varying levels of > success. I am somewhat ambivalent whether I configure it command line or > qmon. > I have been reading the docs online, with some success. > > The two that are currently escaping me: > > 1. I am trying to configure the scheduling. I want a weighted round robin. > Currently Default assigns all jobs to a server until full. Then it goes to > next server. So I might have 24 jobs on each of 3 servers and the other 65 > servers idle. > I have 960 cores, and will be adding another 1000 some odd cores (4 cores > per socket, 2 sockets per server) on about 100 growing to 200 servers. > I want it to set it to send 1 job per server, until all servers have received > a job, then start scheduling a second job on each server.... except when the > jobs on the server are consuming heavy resources, skip to next job. You are submitting the job in a fast sequence? A round-robin like behavior is the default, as the np_load_avg is used to sort the exechosts. When jobs are started fast, the load report they got does not reflect the actual usage yet. You can either add some artificial load in the scheduler (setting "job_load_adjustments" and "load_adjustment_decay_time") or use: https://blogs.oracle.com/sgrell/entry/grid_engine_scheduler_hacks_least > 2. I have a "high cpu" queue, And this is running only on certain machines/blades which are set in the queue's "hostlist"? > where I want it to only be able to send max 1 job per blade. As too many > jobs on one server where it consumes 100% of CPU for 2-4 weeks, will cause > the chips to change speeds. > When I try to configure in qmon, it looks like it wants me to set this on > each blade. Can I set it to the queue in one setting? Setting "slots=1" here will define it at queue instance level. i.e. per machine and you are done. -- Reuti > Thanks in advance > > -- > Dan Hyatt > Division of Statistical Genomics > Washington University School of Medicine > 4444 Forest Park Blvd, Campus Box 8506 > St. Louis, MO 63108 > 314 747 4767 (o) > 314 473 8713 (c) > [email protected] > > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
