[GE users] Using cycles from a 2nd SGE cluster

craffi dag at sonsorol.org
Fri Jan 8 13:39:48 GMT 2010


rhierlmeier wrote:
> However for problems like different funding or fundamental different
> configuration of the clusters I do not see any reason why multi clustering
> should not work.

Just being contrarian here so take my words with a grain of salt !

Multi-clustering works from the perspective of sysadmins, HPC designers 
and senior management.

Few of these people bother to ask users what they think either before or 
after considering multi-clustering.

Few of these people will also measure their cluster from the perspective 
of "how easy is it for me to get my work done?" or "how how many hours 
does my expensive scientist have to spend making her jobs abstracted, 
wrapped and generic so they can be portable?

HPC system operators and management are more concerned with uptime and 
utilization measurements rather than how effectively these systems can 
be used to achieve the organizational goals that created them.

If you ask users what they think about multi-clustering you'd hear:

- dealing with user account UID/GID mapping sucks

- monitoring state and status sucks

- dealing with different filesystem paths sucks

- dealing with data movement issues among clusters sucks

- spending days or hours making simple jobscripts portable sucks when 
that time could be better spent getting actual work done

Unless an organization has static workflows that can be wrapped and 
abstracted by a dedicated app team or unless an organization has 
dedicated sysadmins with domain expertise available to help users with 
optimizing their workflows for cluster portability you will often find 
either unhappy users or users who have given up productivity in order to 
make their own workscripts abstracted and generic enough to be portable. 
For people who have not thought this out carefully you can have a net 
loss in productivity and efficiency.

Again my views are tainted by the industry I work in and the 
organizations I visit. I'm speaking only for myself, not trying to 
describe the entire community or anything.

-Chris

------------------------------------------------------
http://gridengine.sunsource.net/ds/viewMessage.do?dsForumId=38&dsMessageId=237371

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



More information about the gridengine-users mailing list