HP XC System Software Administration Guide Version 2.1

Table 11-2: SLURM Partition Characteristics (cont.)
Characteristic
Description
MaxTime
Specifies the maximum time limit (in minutes) allowed for
jobs in this partition.
The default is unlimited or -1.
MinNodes
Specifies the minimum number of nodes that may be
allocated to any single job.
The default is 1.
Shared
A text string that indicates whether node sharing for jobs
is allowed.
The string va l ues are:
YES
The node may be shared or not, depending
on the a llocation.
FORCE
The node is always available to be shared.
NO
The node is never available to be shared.
State
The state of the partition.
The possible values are UP or DOWN.
Consider a system that h as 12 8 nod es. The following lin e in the /hptc_clus-
ter/slurm/etc/slurm.conf file indicates that partition lsf controls all 1 28 nod es:
PartitionName=lsf RootOnly=yes Shared=FORCE Nodes=n[1-128]
However, you might prefer not to run j obs on the head node, n128. Simply modif y th e line to
the fo llowing:
PartitionName=lsf RootOnly=yes Shared=FORCE Nodes=n[1-127]
Consider an academic system with 256 nodes. Suppose you would like to allocate half the
system for faculty use, and half for student use. Furthermore, the faculty prefers the order and
control i mp osed by LSF, while the students prefer to use the srun command. You might
set up your parti tions as follo ws:
PartitionName=lsf RootOnly=yes Shared=Force Nodes=n[1-128]
PartitionName=cs Default=YES Shared=YES Nodes=n[129-256]
11.3 Restricting User Access to Nodes
Although fu ll u ser authentication is required on every node so that SLURM can launch jobs
and although this access is beneficial for users who need to debug their applications, it can be
a problem because a user could adversely affect the perform ance of another user’s job: they
could log onto any compute node and steal CPU cycles from any job running on that no de. The
solution is to restrict a user to acces s only the nodes that they reserve.
A Pluggable Auth enticatio n Module for use with SLURM (pam_slurm) is supplied with
HP XC System Software for this p urpose. Th is module is designed to work within the
Linux Pluggable Authentication Module (PAM) framework and checks w ith SLURM before
authorizing user access to the local node. Ideally, this should be done as part of the H P XC
system instal lati on.
The pam_slurm module is disabled by default, b ut it can be enabled to restrict the use of a
particular node to only one reserved user at a time. Using this m od ule can stop users from
adversely affecting another user’s job. Before you enable this module, you must have a Login
role d efined on at least one node.
11-6 SLURM Administration