Am 25.04.2013 um 00:19 schrieb Dave Love: > Reuti <[email protected]> writes: > >> Some time ago Fritz mentioned that SCM to control it is abandoned and >> Univa integrated something new to control it. > > SDM is still available if you want it, and the others I'm aware of are > on <http://arc.liv.ac.uk/SGE/tools.html>, apart from one funded with my > taxes which I could never get hold of. (They could do with assessing > and comparing, if someone would like to contribute. It would also be > interesting to know how they compare with support in SLURM and OAR, for > instance.) > >> Nevertheless there are scripts like >> http://kimura.univ-montp2.fr/redmine/projects/cluster-tools/wiki/Greenginecode >> making a little bit of looking ahead. It's necessary to replace their >> calls to shutdown/start (originally ssh/WOL) with calls to ipmitool >> though to control it in a nicer way (Link "Repository" in the top row, >> other links are using a wrong port to download it). > > [I'd use freeipmi, or powerman to be more general.] > >> What's not working is advance reservation. > > Do you mean it takes resource requests and their reservation into > account properly?
No, not directly. It scans the output of `qstat` of any given queue for waiting jobs (hence: which may run in this queue) and bases the decision on this. > I doubt this job can be done well without support in SGE itself > (e.g. scheduling as if "zombie" nodes were available). > > Note that power cycling may hurt reliability (see previous discussion > here). Running with the Linux "ondemand" frequency governor > significantly reduces consumption of idle nodes without significant > problems we've seen. If necessary (latency effects?) you could flip > between performance and ondemand in the usual SGE hooks. I second this. But I saw locations where the cpufreq daemon wasn't installed and they chose the above solution instead. -- Reuti _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
