[gridengine users] Centos 7 and Gridmaster install
reuti at staff.uni-marburg.de
Wed Jan 4 18:22:17 UTC 2017
Am 14.12.2016 um 18:31 schrieb Christopher Black:
> We’ve done hundreds of compute nodes with NFS, but keep spools and executables local and only $SGE_ROOT/$SGE_CELL/common on NFS. Depending on nfs server implementation you may need to do some nfsd and mount option tweaking.
> This may be worth reading: http://gridscheduler.sourceforge.net/howto/nfsreduce.html
I also made some experiments in the past about it, but didn't find the saved time of shared executables compared to local ones worth to bother any more.
To avoid the update of the local executables, I mounted first /usr/sge to somewhere on the nodes and copied the content to a local loop device. After unmount of /usr/sge the image could be mounted thereon and only the /usr/sge/default/common I mounted then inside the mounted loop device. So I got fresh copies of SGE with every reboot of an exechost.
But I couldn't see any great improvement compared to a complete NFS share of /usr/sge.
More information about the users