[Moabusers] nodeindex question

[Douglas Wightman]

The NODEINDEX is something that moab can use to speed up node operations
on large systems.  Some systems will name their nodes based on their
location in the cluster (rack and slot).  Moab can be configured to
understand this.

It's definitely not necessary for normal and correct operation.  It is
just a feature that can be used on very large systems (1000+ nodes) to
help Moab speed things up internally.

In your specific case, Moab saw the "1", "2", "3", and "4" on the ends
of the names and interpreted those as the index.  Because "reliant" has
no number at all Moab could not interpret any node index.  This will not
have any side effects whatsoever.

Hopefully that answers your question,

- Douglas

On Wed, 2007-09-26 at 20:52 -0700, Steven DuChene wrote:
> I have a small moab/torque setup for testing where four of the available 
> compute resources are
> x86_64 quad core nodes in an Oscar cluster. Another compute resource is 
> a 8-way ia64 shared
> memory system that is also running a torque mom daemon and is listed in 
> the torque server's nodes file.
> 
> When I run mdiag -n -v on the headnode I get the following output back:
> 
> compute node summary
> Name                    State   Procs      Memory         Disk          Swap      Speed   Opsys   Arch Par   Load Rsv Classes                        Network                        Features              
> 
> oscarnode1               Idle    4:4     16013:16013       1:1       16424:16424   1.00   linux      - bas   0.00   0 [batchx86_4:4]                 [DEFAULT]                      x86_64               NODEINDEX=1
> oscarnode2               Idle    4:4     16013:16013       1:1       16425:16425   1.00   linux      - bas   0.14   0 [batchx86_4:4]                 [DEFAULT]                      x86_64               NODEINDEX=2
> oscarnode3               Idle    4:4     16013:16013       1:1       16425:16425   1.00   linux      - bas   0.00   0 [batchx86_4:4]                 [DEFAULT]                      x86_64               NODEINDEX=3
> oscarnode4               Idle    4:4     16013:16013       1:1       16425:16425   1.00   linux      - bas   0.00   0 [batchx86_4:4]                 [DEFAULT]                      x86_64               NODEINDEX=4
> reliant                  Idle    8:8     31380:31380       1:1       78575:78575   1.00   linux      - bas   0.00   0 [batchia64_8:8]                [DEFAULT]                      ia64                
> -----                     ---   24:24    95432:95432       5:5      144274:144274
> 
> Total Nodes: 5  (Active: 0  Idle: 5  Down: 0)
> 
> 
> Notice that over in the very last column of the output there are 
> "NODEINDEX" entries for the four nodes of the distributed
> Oscar cluster but no nodeindex for the ia64 shared memory system. Why is 
> that? According to the documentation I have
> found so far in section 5.2 of the MWM guide:
> 
> NODEINDEX     node's nodeindex as specified by the resource manager 
> 
> If that is true why hasn't torque specified a nodeindex for the shared 
> memory system?
> --
> Steve DuChene
> _______________________________________________
> moabusers mailing list
> moabusers at supercluster.org
> http://www.supercluster.org/mailman/listinfo/moabusers