[GE users] Several beginner questions. Global Q does not see any slots.

templedf dan.templeton at sun.com
Mon Mar 16 13:37:35 GMT 2009


muxecoid wrote:
> Hello. I'm completely new to building clusters, please, try to give me a "for dummies" explanation.
>
> 1. Do I need to use a special distribution to build a grid? Can I just connect a few computers with vanilla ubuntu with network switch, make a network mounted directory and use them?
>   

That should be fine.

> 2. Do I need to install sge-root locally on hard disk for every execute node?
>   

No.  The SGE root can be on a shared file system.  In large clusters 
it's not recommended, but "large" here means several hundred nodes.

> 3. Can I run execution server on master machine? The cluster currently consists of only 3 machines now and will expand to 8 later so making a dedicated master is a waste.
>   

Yes, and for such a small cluster, that's probably a good idea.

> 3.1. Before installing on the actual computers I installed it on local desktop. I used the same machine for both master and execution install. I used qmon to configure it. I created a new user group, added a user to it, than added the user to list of users, than added the execution host, than did install_exec, added the user group to allowed for both host and queue.
>
> On cluster queue screen I see "USED:0 AVAIL:0". On the host screens I actually see the host and the real load values (I loaded it by running stuff directly). What did I do wrong?
>   

Check the host list for the queue.  Is your one host listed?  If not, 
add it.

> 4. Does SGE know to balance load wisely on machines with hyperthreaded CPU?
>   

SGE sees CPU threads as cores and treats them as such.  (Someone correct 
me if I'm wrong here...)  It is not aware of how threads may contend 
with each other.  It is possible to configure SGE to use processor sets 
or bind jobs to cores, though, so you may be able to get the desired 
effect manually.  SGE also has a flexible notion of "slots."  By 
default, SGE assumes one slot per core per host, but that's completely 
arbitrary.  If it's better for your machines to run fewer jobs than 
that, then set the number of slots to a smaller number.

Daniel

> ------------------------------------------------------
> http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=133344
>
> To unsubscribe from this discussion, e-mail: [users-unsubscribe at gridengine.sunsource.net].
>

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=133364

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



More information about the gridengine-users mailing list