Grid Account

In order to obtain a grid account, please send an email to Steve Spencer containing your full name, Mac ID, Student Number and supervisor name. Your password will initially be your student number (with no leading zeros) but you can change it by logging in and typing

passwd

Background

NOTE: Files stored on grid are NOT backed up so it is in your best interests to ensure that you have copies of your important files stored elsewhere as well as on grid.

The Blade Center consists of 140 compute nodes capable of running Matlab 7.1. They are all named gridxxx where xxx = 001 through 140. Their hardware configuration is as follows:

· grid001 - grid004 - 4 nodes with dual 3.33 GHz Dual Core Xeon (X5260) CPUs and 6 GB of memory

· grid005 - grid014 - 10 nodes with dual 2.4 GHz Pentium 4 Xeon CPUs and 8 GB of memory

· grid015 - grid064 - 50 nodes with dual 2.4 GHz Pentium 4 Xeon CPUs and 1 GB of memory

· grid065 - grid069 - 5 nodes with dual 2.6 GHz Pentium 4 Xeon CPUs and 1 GB of memory

· grid070 - 1 node with dual 2.6 GHz Pentium 4 Xeon CPUs and 4 GB of memory

· grid071 - grid084 - 14 nodes with dual 2.0 GHz dual core Opteron CPUs and 2 GB of memory

· grid085 - grid094 - 10 nodes with dual 2.2 GHz dual core Opteron CPUs and 4 GB of memory

· grid095 - grid098 - 4 nodes with dual 2.2 GHz dual core Opteron CPUs and 8 GB of memory

· grid099 - grid140 - 42 nodes with dual 3.06 GHz Pentium 4 Xeon CPUs and 1 GB of memory

Sun Grid Engine is installed on an external management system to distribute users across the compute nodes in an efficient manner. Users can only connect directly to the management system which then schedules their task on one of the compute nodes. Users' files reside on an external file server and are accessible on the management system and on every compute node.

To Access the Blade Center

ssh grid.ece.mcmaster.ca

If you don't have ssh, we can install a copy on your PC. Once connected you will be using bash by default and if you want to view graphics then you need to identify your Xserver, eg.

export DISPLAY=your-computer:0

To submit batch jobs

qsub your-batch-script

For a sample batch script see /usr/local/sge/examples/jobs/simple.sh on grid

To Run an Interactive Session

If you are using a Windows PC, make sure to start X-Win32 or Cygwin on your PC first. If you don't have X-Win32 or Cygwin, we can install it on your PC.

If you are using a Unix system or Cygwin, you may need to allow other systems to connect to your Xserver by typing

xhost grid

Once you have an X-windows environment up and running, open a shell and type

qsh

which will start an xterm on your display, connected to one of the grid nodes.

NOTE: DO NOT start more than 1 job in any interactive shell because this circumvents the load manager to the eventual detriment of all.

It is assumed that if you use qsh then your intention is to run a task where you interact with an application. If you do not intend to interact with your application then use qsub.

Matlab can be run in either mode depending on your requirements. DO NOT run anything locally on grid, use qsh, qsub or qmon to schedule a task on one of the compute nodes.

Monitoring your Job(s)

To monitor or schedule jobs of any type you can use the graphical interface

qmon&

Other Documentation

For more information about these commands,

man qsub

man qsh

man qmon

For more information about the load manager, grid engine, and the various commands see

/usr/local/sge/doc/SGE53AdminUserDoc.pdf

on grid.ece.mcmaster.ca. Since there is intentionally no Acrobat reader on grid you will have to use sftp to copy this file to your desktop system for viewing.