[gridengine users] Consumable configuration best practices question for hundreds of resources for specific group of nodes

William Hay w.hay at ucl.ac.uk
Thu Apr 9 19:01:08 UTC 2015

On Tue, 7 Apr 2015 14:08:57 +0000
Yuri Burmachenko <yuribu at mellanox.com> wrote:

> >*You could define your queue configuration so that all queues on all
> >other nodes have 0 of the resource available while the nodes with
> >access say nothing about availability and therefore have access to
> >the full resources defined on the global host.
> >*You could define the resources as having 0 availability on hosts
> >other than the ones in question.
> >*You could probably also do the same with resource quotas.
> >The first of the above is probably simplest/least work assuming your
> >existing queue configuration is simple.
> Will the first approach require updating of all execution hosts
> configuration (qconf -se)?
No.  You configure the cluster queues which apply to all the queue
If you just have an all.q setting it's complex values something like

complex_values h_vmem=16G,cell1=0,cell2=0,cell3=0,[@magic=h_vmem=16G]

combined with defining complex_values on the global to declare 1 of
each available cell and creating the @magic host group with all the
hosts connected to your magic wonder gadget.  

h_vmem is a standin for other resources you want defined.

> Maybe it will be simpler to install a dedicated SGE scheduler for
> this task, what do you think?
It might be marginally simpler to configure but might make the
workflows of the people using the cluster more complex.  Your choice
but I'd go for the one off config complexity. 

> Also instead is it possible to install it as a separate cluster by
> reusing existing SGE_ROOT - is it recommended?

It's not exactly common practice but if you want two clusters no reason
it shouldn't work.
> >> All cells are different and users will need to know which one they
> >> need to request. At this stage they all should be distinct.
> >OK.  If users request a lot of different cells for individual jobs
> >this will probably lead to long delays before jobs start.  Said
> >users will almost certainly want to request a dynamic reservation
> >for their jobs (-R y).
> Can you explain how dynamic reservation will help?

If users request lots of different and highly specific cells they have
to wait until they are all free simultaneously.  Jobs which only
request a few cells will slip in ahead of higher priority jobs that
request a lot of different cells unless the jobs that want a lot of
cells can reserve those cells that are free until all the cells they
need are free.

Without reservations the jobs that need a lot of cells might find there
is never a time when all the cells they need are free.

The scheduler will allow short jobs that require reserved cells to use
them provided they won't make the worst case delay for a higher
priority job with a reservation worse.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://gridengine.org/pipermail/users/attachments/20150409/3e6b2c3c/attachment.sig>

More information about the users mailing list