Difference between revisions of "Son of Gridengine"

From wiki
Jump to: navigation, search
(Install the Son of Gridware RPMs)
(chown -R sgeadmin.gridware sge)
Line 49: Line 49:
  
 
== chown -R sgeadmin.gridware sge ==
 
== chown -R sgeadmin.gridware sge ==
 +
 +
 +
==edit /etc/bashrc==
 +
 +
Add the following two lines to /etc/bashrc
 +
 +
 +
SGE_ROOT=/opt/sge; export SGE_ROOT;
 +
PATH=/opt/sge/bin/linux-x64:$PATH
  
 
= Administration =
 
= Administration =

Revision as of 10:36, 11 March 2019

Introduction

After Oracle bought Sun and then shut-sourced the Sun Grid Engine, Open Grid Engine made a release based SGEv5 in 2012 which hey called GE2011. However there were no further releases. Then ARC at the University of Liverpool started releasing its "Son of Gridengine and have been maintaining updates to it at least as far March 2016.

Until September 2016, the Queue manager in the marvin cluster was GE2011 which was getting a bit old, so when the queue manager failed due to a corrupted database

Steps

Administrative host setup

All nodes must be set up as administrative hosts, despite the fact that only the master seems to be "administrative"


Getting the XML::Simple perl module

The RPMForge Extra repository are need for this. These can be installed via an RPM, and afterwards the Extra branch of the repo much be enabled as it is not enabled by default.

Note that it is best to disable this after all the RPMs have been installed


Install the Son of Gridware RPMs

Centos 7 requires the epel repo installed

yum install https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm

And the following packages:

yum install jemalloc-3.6.0 lesstif-0.95.2 munge-libs-0.5.11 libdb4-utils-4.8.30


yum install -y gridengine-8.1.9-1.el6.x86_64.rpm gridengine-devel-8.1.9-1.el6.noarch.rpm gridengine-execd-8.1.9-1.el6.x86_64.rpm gridengine-qmaster-8.1.9-1.el6.x86_64.rpm gridengine-qmon-8.1.9-1.el6.x86_64.rpm


From https://arc.liv.ac.uk/downloads/SGE/releases/8.1.9/

For a worker node this is rather excessive, but such is the nature of the binary-chek stage of the "install_exec" script that all of these are necessary.

copying the default/common directory over to the node

First the default directroy must be created:

ssh nodeX 'mkdir /opt/sge/default'

And then followed by:

scp -r common node8:/opt/sge/default


chown -R sgeadmin.gridware sge

edit /etc/bashrc

Add the following two lines to /etc/bashrc


SGE_ROOT=/opt/sge; export SGE_ROOT;
PATH=/opt/sge/bin/linux-x64:$PATH

Administration

Creating a new parallel environment

  • Copy out an current parallel envioment out to a file
  • edit this file as you wish
  • execute
qconf -Ap <my_pe_file>

A crucial oversight is to forget that this new parallel environment needs to be inserted into the queue's configration.

Queues and hostgroups

the dohfq.sh script accepts a rootname and list of numbers. The rootname becomes @rootname hostgroup and rootname.q for the queue. Node0 is in fact marvin. 1, is node 1 etc. These are the nodes to be ncluded in the new queue.