Choosing more than one architecture type

Blog post edited by Anonymous - "Migration of unmigrated content due to installation of a new plugin"

Although all of our nodes are binary compatible, the HPCC is a heterogeneous cluster based built from many different types of nodes. The node types are based on the type of architecture that the node is built on and the year the nodes where purchased. Here is a current list:

Cluster name

|

Development node

|

Chip Manufacturer

|

Year Purchased

|

Node Prefix

---|---|---|---|---

amd05

|

dev-amd05

|

AMD

|

2005

|

amd

intel07

|

dev-intel07

|

Intel

|

2007

|

scw

amd09

|

dev-amd09

|

AMD

|

2009

|

afn

gfx10

|

dev-gfx10

|

Intel with nVidia Graphics Card

|

2010

|

nvx

intel10

|

dev-intel10

|

Intel

|

2010

|

icx

intel11

|

n/a

|

intel

|

2011

|

ifi

You can specify a specific architechture in your submission script using feature= command. For example the following will only run on the intel10 node:

1
2
3
4
5
6
7

#!/bin/bash --login
#PBS -l nodes=1:ppn=1,mem=1gb,feature=intel10,walltime=03:59:00
#PBS -j oe

cd ${PBS_O_WORKDIR}

#Your stuff here...

You can also specifiy multipule architechtures using the PIPE (OR) charicture. For example the following will run on the intel07 or the intel10 nodes:

1
2
3
4
5
6
7

#!/bin/bash --login
#PBS -l nodes=1:ppn=1,mem=500mb,feature=intel07|intel10,walltime=00:05:00
#PBS -j oe

cd ${PBS_O_WORKDIR}

#Your stuff here...

Unfortunately, as far as I can tell, the gfx10 nodes will not work using the OR syntax. This is probably because the gfx10 nodes also use a different network fabric (gbe) and I have not figured out the way use more complex logic with the feature= resource.

View Online

Blogpost migrated from ICER Wiki using custom python script. Comment on errors below.