[GE users] SGE noobie

craffi dag at sonsorol.org
Tue May 26 20:55:23 BST 2009

On May 26, 2009, at 3:11 PM, biostat wrote:

> Thanks alot for your help so far. I feel like I am making headway  
> (albeit slow).
> Here is, I think, where I start to get tripped up:
> 1.Install qmaster (using ./install_qmaster) on machine1:/SGE6.

Sounds good

> 2. NFS machine1:/SGE6 so that it is visible by machine2.

Sounds good. You want to make sure you are not disabling setuid  
operations or remapping the root user to a non-privledged UID. There  
are some very strict NFS export permissions that can break the SGE  
install. You need to install as root initially (can run as non-root  
user after install) and there are setuid binaries within SGE that must  
be able to function for the grid to be used by more than one user.
> 3. Install execd (using ./install_execd) on machine2. When I install  
> execd here, should I be running the install_execd script that is  
> locally on machine2 (lets say under the directory machine2:/ 
> SGE6b  ), or should I be running the install_execd script that is on  
> machine1:/SGE6?

General rule: run the script from the machine where you want the  
software to be installed

In your case I'd run the install_execd script from both machines. That  
will allow both machines to have execd's installed and both will be  
able to do work in the cluster.

> 3. When the execd install asks for my SGE_ROOT var, should I give it  
> machine1:/SGE6 or machine2:/SGE6b?

Your SGE root will be "/SGE6"  -- SGE wants the local file path,  
telling it NFS semantics like machine1:/SGE6 would confuse it. Mount  
first via NFS and then your SGE_ROOT becomes "/SGE6" everywhere

> I guess the essence of my question is: at what points should in the  
> install should I be working with the remote directory, and at what  
> points should I be working with the local directory?

Run the scripts from the machine where you want the software to be  
installed is the general rule.

> Also, as a possibly related question, when the qmaster install asks  
> for a list of admin and submit hosts, should I give it all the hosts  
> in my cluster, including itself (assuming it is the only qmaster,  
> and there are 1 or more other machines that are all execution hosts)?

Harmless - you can easily add/remove these hosts after the fact. Just  
name machine2 when you install the qmaster so that the install of  
execd works. There are some parts of the install_execd script that  
want the node to be listed as an admin host so that it can bootstrap  
into the default all.q cluster queue that gets set up.

> Thanks again, your help is really appreciated.
> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=199033
> 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