[GE users] How important is checkpointing support?

dangruhn Dan.Gruhn at groupw.com
Wed May 20 16:30:21 BST 2009


I've been working on something else so sorry for the delay:

agay wrote:
> Dan, thanks a lot! It sounds like an elegant solution! 
>> So far so good, 
> I have very little experience with virtual machines so I'm asking dumb questions. Are your Windows users satisfied with the VM? Do you authenticate them in Linux before starting the VM? Do each user gets a "fresh" VM so viruses can't pass between consecutive user logins?
Our test user is very satisfied with the VM. I talked to her this 
morning and she didn't even know she was using a virtual machine.  She 
is probably a fair example of a typical office worker-type user.

She does authenticate in Linux (Fedora 9) and then the X startup script 
automatically drops her into VMware Player and on into Windows XP. We 
did have to install an extra bit of software to get the X startup 
scripts running, this doesn't happen by default in Fedora 9.

We do not give a fresh VM after each logout.  This would (I think) 
require Windows to be shutdown and VMware to be restarted.  In our setup 
it works the same way a non-virtual PC would work: one user logs out and 
the next  logs in.  The only difference the user sees is the small 
VMware player menu bar on the top of the screen which can be made to 
auto hide when not in use.
> Our situation is probably shared by other educational institutions. There are several large Windows farms serving as student labs that we wish to use as night time compute clusters. We must provide the Windows environment during the day with a lot of applications and make sure each user gets it clean.
It sounds like this approach would be ideal for you.  You'll need to do 
some things we aren't, but I don't see any show stoppers.
> The SGE docs may give an impression that checkpointing is an easy to use panacea for load balancing but this seems inaccurate. With the Windows running in a VM you can keep crunching numbers and needs checkpointing only if there are really long jobs.
To be fair, SGE seems to support checkpoint/restart quite nicely.  The 
problems are really at the application level: applications that can 
easily stop running on one machine and pick-up where it left off on 
another machine are few and far between.
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=196027
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].


To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].

More information about the gridengine-users mailing list