On Thu, May 24, 2012 at 5:44 AM, William Hay <[email protected]> wrote:
> I assume you are referring to your cgroups stuff here.  Some of us are
> still using Linux versions without cgroups
> (like Scientific Linux 5).

William, yes and no... I was mainly referring to cgroups integration's
ability to handle process tagging, but we also have something else
that we are adding in GE 2011.11u2 if most of our users are still
using very old kernels - note that the first cgroups capable kernel
came out in May 2008.

BTW, our ssh HOWTO was updated yesterday to reflect the cgroups
integration's ability to use a stock sshd for tight job control:

http://gridscheduler.sourceforge.net/howto/qrsh_qlogin_ssh.html


For those who are still using RHEL 5, CentOS 5, Oracle Linux 5, or
Scientific Linux 5 - ie. all the common RHEL based Linux
Distributions, a kernel upgrade is needed to use cgroups. It may sound
like a lot of work, but Oracle has already created a yum installable
package for the all of the RHEL based distributions. Installing
Oracle's Unbreakable Enterprise Kernel is as simple as upgrading a
package, and UEK is compatible with RHEL for the user-space
applications point of view. In fact, Oracle DB & middleware, SAP are
certified on both the RHEL based kernel and Oracle's UEK.

See SAP's announcement: http://scn.sap.com/community/oracle

We (Open Grid Scheduler) tested our code on both the RHEL based kernel
and Oracle's UEK starting from the hwloc integration back in April
2011:

http://gridscheduler.sourceforge.net/projects/hwloc/GridEnginehwloc.html

As a newer kernel, the UEK has much better performance than the old
RHEL 5 based kernel. Note that UEK 2 was released earlier this year
and has more performance improvements & new features - eg. BrtFS. The
only catch is that if you have kernel modules that require specific
kernel versions and can only use a 4+ year old kernel, then let us
know and if there are enough number of sites running old kernels, we
will add another new capability in PDC to track the stock sshd and
handle it as if it is a tight SGE-SSH daemon.



> With cgroups making for a much more enforceable resource
> control mechanism could this be relaxed?

Thanks again for the suggestion - we will look into it. If it is
technically doable, we will add it to GE 2011.11u2 the latest.

Rayson



>
> William
_______________________________________________
users mailing list
[email protected]
https://gridengine.org/mailman/listinfo/users

Reply via email to