Am 08.11.2012 um 09:01 schrieb Arnau Bria: > I'd like to configure my nodes in the way that if some node reaches X > load value, it does not accept more jobs. > > The only way I've found for doing so is defining queue level load_avg > and create some host group and then assign load value to host group: > > # qconf -sq cn-el6|grep ^load > load_avg=1.75,[@cn=np_load_avg=1.25]
You can also use exechost names in this syntax without the @: load_avg=1.75,[node001=np_load_avg=1.25],[node002=np_load_avg=1.30] But: the "load" does nowadays no longer reflect only busy cores, but also process in unterruptible sleep. Load can go high even with idling cores. Unless you use the nodes also outside of SGE, it's better to focus on the available and granted slots and don't define any load_thresholds, which should be set to the number of cores. In our environment also with disabled HT (hyper threading) just the plain number of cores. -- Reuti > but, is there any way for configuring it at host level without having > to create one host group for each node? > > PS: I'm reading "OGE Administration Guide" but I've not seen how to do > so. The guide describes many commands but not the params or gives some > examples. Is there any other administartor guide that I could read? > > gridengine-2011.11p1-2.el6.x86_64 > > TIA, > Arnau > _______________________________________________ > users mailing list > users@gridengine.org > https://gridengine.org/mailman/listinfo/users _______________________________________________ users mailing list users@gridengine.org https://gridengine.org/mailman/listinfo/users
