[scyld-users] Re: diskless nodes

[Donald Becker]

On Thu, 14 Nov 2002, Chris Lipp wrote:

> I am a newbie to clustering and looking for some assistance. I have set
> up a small beowulf cluster (10 slave nodes connected to the front-end
> with a 100Mb/s switch) using the SCYLD release.

Which release?

> I have had no problems
> bringing the system up online and booting all of my diskless nodes, and
> recieving the "up" status signal from the beosetup monitor. 
> 
> My problem arises when I go to run a benchmark; every benchmark I try doesnt
> seem to be using the nodes. I have tried many while monitoring the nodes
> in the beostatus window with no response. 

Try using
  export NP=10
or
  export ALL_CPUS=1

> The only thing I have been able to determine independently is that the
> slave nodes are not reporting any memory during boot-up (I have
> upgraded the bios - no change) the slave nodes are old HP vectra
> VLs. The beomap function only returns the -1 response from the front
> end. I can bpsh into all the nodes with no problems. 


Test the scheduler by running
  NP=10 beomap

You should get back (on an idle cluster):
  -1:0:1:2:3:4:5:6:7:8

Note that the first process ("rank 0" for MPI) is on the master, and you
use only 9 slaves.

ALL_CPUS=1 beomap
  -1:0:1:2:3:4:5:6:7:8:9

ALL_CPUS=1 NO_LOCAL=1 beomap
  0:1:2:3:4:5:6:7:8:9

-- 
Donald Becker				becker at scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993