You can in fact use daemons or client-side commands taken from an Open Grid Scheduler/Grid Engine tarball and use them in an existing SGE 6.2u5 cluster. For example, if you want to plug the LD_* env. var. security hole:
http://gridscheduler.sourceforge.net/security.html You can run: 1) Stop new job scheduling 2) mv sge_shepherd sge_shepherd.old 3) cp <path>/sge_shepherd . 4) re-enable job scheduling This way, you don't need to stop the whole cluster (and in fact running jobs would continue to run with this method) to install a patch. Also, we have users who run execd & shepherd binaries taken from Open Grid Scheduler/Grid Engine 2011.11 in older SGE 6.2u5 clusters. This is how they took advantage of the hwloc binding library without migrating. (Note that in order to do that, you will need to create a symbolic link to point "lx24-amd64" to "linux-x64" in the bin & lib directories.) Finally, sge_qmaster in Open Grid Scheduler/Grid Engine 2011.11 can be mixed with SGE 6.2u5 as well. You can run in fact run this following to upgrade a live cluster: 1) mv sge_qmaster sge_qmaster.old 2) move in sge_qmaster binary from the Open Grid Scheduler/Grid Engine 2011.11 tarball And later when you do a restart of the master machine (either by accident - like power outage, or you wanted to reboot it after installing OS patches), the new qmaster comes up and is able to talk to SGE 6.2u5 execution hosts. You would then have the option of using NFSv3 for the Berkeley DB spooling as mentioned in our release notes: http://gridscheduler.sourceforge.net/Releases/ReleaseNotesGE2011.11.pdf So what it comes down to is that you have the option to not upgrade the whole cluster all at once. And also, mv is your friend as most operating systems map the executable into memory when they run a program, and this means you will get a seg fault if you use cp to overwrite the older SGE 6.2u5 binaries with the Open Grid Scheduler/Grid Engine binaries. Rayson On Thu, Apr 26, 2012 at 5:18 AM, admin iqtc <[email protected]> wrote: > Hi all, > > on the OGS webpage, it says: >> >> Grid Engine 2011.11 is compatible with Sun Grid Engine 6.2u5. You can even >> run Grid Engine 2011.11 on a subset of nodes while the rest of the cluster >> runs the older 6.2u5 release - ie. as a drop-in replacement for 6.2u5. > > > So, how is that achieved? I've searched on the mailing list and google but i > can't find anything regarding that. Is the key on the init script or > something? > > Thanks > > _______________________________________________ > users mailing list > [email protected] > https://gridengine.org/mailman/listinfo/users > _______________________________________________ users mailing list [email protected] https://gridengine.org/mailman/listinfo/users
