Defining external host-based resources in Platform LSF

This tip describes how to define site-specific external resources on hosts in the cluster for scheduling purposes. LSF supports three types of external resources, boolean, string and numeric. Use the following steps to define and use external host-based resources in LSF.

Note:  You must be an LSF administrator to perform these steps.

1. Define external resources in lsf.shared

You must specify at least a name and a brief description in the RESOURCENAME and DESCRIPTION columns. The following example has three resources of different types. All are static resources. For dynamic resources, you must use INTERVAL to define how frequently the resource is updated.

Begin Resource
RESOURCENAME  TYPE    INTERVAL INCREASING  DESCRIPTION       
   bigmem     Boolean ()       ()          (Hosts with very big memory)
   switch     String  ()       ()          (network switch)
   appA       Numeric ()       N           (Application Licenses)
End Resource

2. Define resources in the RESOURCES column of Host section of lsf.cluster.cluster_name.

Define all host-based resources in the RESOURCES column. Boolean resources use the resource name directly, while numeric and string resources use <resource_name>=<resource_value> format. Here is an example:

Begin   Host
HOSTNAME  model    type        server r1m  mem  swp   RESOURCES   
fatnode   !        !           1      3.5   ()   ()   (bigmem appA=4)
s1node1   !        !           1      3.5   ()   ()   (switch=s1)
s1node2   !        !           1      3.5   ()   ()   (switch=s1)
s1node3   !        !           1      3.5   ()   ()   (switch=s1)
s1node4   !        !           1      3.5   ()   ()   (switch=s1)
s2node1   !        !           1      3.5   ()   ()   (switch=s2)
s2node2   !        !           1      3.5   ()   ()   (switch=s2)
...
s4node1   !        !           1      3.5   ()   ()   (switch=s4)
s4node2   !        !           1      3.5   ()   ()   (switch=s4)
s4node3   !        !           1      3.5   ()   ()   (switch=s4)
s4node4   !        !           1      3.5   ()   ()   (switch=s4)
End     Host

3. Run lsadmin reconfig and badmin mbdrestart to reconfigure LIM and mbatchd to make the new resource definition take effect. Use lshosts to verify the result.

$ lshosts fatnode
HOST_NAME    type    model     cpuf  ncpus  maxmem maxswp server RESOURCES
fatnode      X86_64  Intel_EM  60.0  16     128G   32G    Yes    (bigmem)

$ lshosts -s
RESOURCE                                VALUE       LOCATION
switch                                     s1        s1node1
switch                                     s1        s1node2
switch                                     s1        s1node3
switch                                     s1        s1node4
...
switch                                     s4        s4node4
appA                                        4        fatnode

Users can now use those resources in job resource requirement. Here are a few examples.

The following job asks for 4 slots and runs on a fatnode host:

• bsub –n 4 –R “select[fatnode] span[hosts=1]” myjob

The following job must run within same switch and asks for 8 slots:

• bsub –n 8 –R “same[switch]” myjob