[gridengine users] Spooling for a shadow master...
j.abbott at imperial.ac.uk
Thu Mar 31 09:23:33 UTC 2011
Hi Joachim, Stephen,
Many thanks for the quick responses. I'll stick with the classic
spooling for now then...spooling performance is the least of our worries
since we only talking about 25 hosts.
On Wed, Mar 30, 2011 at 03:49:11PM +0100, Joachim Gabler wrote:
> Hi James,
> there are the following spooling options if you want to setup sge_shadowd:
> * classic spooling on nfs (or nfs4)
> * Berkeley DB spooling on nfs4
> * Berkeley DB RPC server (still available in Grid Engine 6.2u5, but no
> longer supported with Univa Grid Engine 8.0.0)
> With your current setup you can install sge_shadowd without changing your
> spooling setup.
> In deployments with a large number of hosts, many queues and esp. high job
> throughput Berkeley DB spooling will give you better performance.
> But switching the spooling method can not be done on the fly - this would
> require re-installation of your cluster (could be done via inst_sge -upd).
> Regarding NFS4 on Linux: NFS4 support on older Linux versions is pretty
> broken. But with current Linux versions NFS4 seems to be OK. We are
> currently testing Berkeley DB spooling on NFS4 on Linux, so far with good
> In your case I would recommend sticking with classic spooling for now,
> should you experience performance issues due to spooling, you might want
> to switch to Berkeley DB.
> Classic spooling should provide sufficient security (in has code to
> prevent data corruption due to filesystem write errors), but Berkeley DB
> will certainly provide higher security - it provides ACID (atomicity,
> consistency, isolation, durability) criteria, see
> Best regards,
> James Abbott wrote:
> I'm in the process of reorganising our SGE setup, and would appreciate
> some guidance on a couple of points re: spooling.
> Our $SGE_ROOT is exported to all hosts from a separate clustered, highly
> available NFS4 server. The qmaster is using classic spooling to
> SGE_ROOT/spool/qmaster. I want to add a shadow master for improved
> resilance, but have got a little confused over the 'best' way of
> handling spooling for this setup.
> I understood it was necessary to use a BDB spooling server if using
> shadow masters, but see that this is essentially being deprecated due to
> BDB RPC issues. I've also read that classic spooling is ok via nfs4
> (hence our current setup...), but have since seen comments that this
> only applies if the qmaster is running on Solaris (our SGE
> infrastructure is all CentOS 5.5).
> So, adding a BDB spooling server introduces RPC security weaknesses and
> a single point of failure, and looks like will not be an option in
> future releases so doesn't look like an ideal solution. The NFS server
> currently hosting SGE_ROOT is a clustered, highly available system, so
> gives us resiliance, but is using NFS4 for classic spooling really safe?
> It seems to be working ok for us with a single qmaster, but that may be
> due to luck...
> Bottom line question: What is the optimal spooling setup for using
> shadow masters?
> If we do decide to move to a BDB server, can I setup the spooling server
> and simply shutdown the qmaster, edit
> $SGE_ROOT/$SGE_CELL/common/bootstrap and restart the qmaster, or is it
> more involved than that?
> Many thanks,
> Notice from Univa Postmaster:
> This email message is for the sole use of the intended recipient(s) and
> may contain confidential and privileged information. Any unauthorized
> review, use, disclosure or distribution is prohibited. If you are not the
> intended recipient, please contact the sender by reply email and destroy
> all copies of the original message. This message has been content scanned
> by the Univa Mail system.
> users mailing list
> users at gridengine.org
Dr. James Abbott
Bioinformatics Software Developer
Imperial College, London
More information about the users