Monash Sun Grid (MSG)

• Service Summary 
• Service Statement  

The Monash Sun Grid is a high-performance compute facility ideally suited to development of tasks prior to dispatch to VPAC/APAC, or for tasks that those external compute facilities are not well suited to, in particular data-intensive tasks. It is a component of the Monash Campus Grid.

The Monash Sun Grid was commissioned in April 2005 to meet the needs of high-end compute researchers within Monash.

Overview of current capabilities:

• GNU Compilers: C, C++, Fortran 95
• Intel Compiler: Fortran 95
• Batch System: Sun N1 Grid Engine 6.1
• Grid Services: Globus Toolkit 4.0.5
• Dual gigabit ethernet connection to each node
• Head node with failover capability
• Applications Software OpenMPI, NAMD, CCP4, CTSIM, NetCDF, WRF, FLTK

Load utilisation figures can be seen at http://msg.its.monash.edu.au/ganglia/

Click here to access User Guide

MSG Execute Nodes

The MSG now provides three classes of execute node.  All are currently being expanded.

• Standard cores (MSG-I, MSG-II, MSG-III):  288 CPU cores with 2 GB RAM per core.
• Hi-RAM cores (MSG-IIIe):  96 CPU cores with 10 GB RAM per core (40 GB RAM per node).  These can be flexibly configured in a variety of ways to meet specific high memory requirements.  Please contact Philip Chan  to configure the MSG-IIIe slots to meet specific memory resource requirements of your jobs.  Some examples of possible memory configurations are shown in Table 3 below.
• Tightly-coupled cores (MSG-IV aka ‘Brecca-in-a-box’) for communication-intensive MPI applications and larger-scale (up to 32-process) shared-memory applications:  96 CPU cores at 3 GB RAM per core (96 GB RAM per node) with high-bandwidth hyper-transport and Infiniband CPU/memory interconnects.  Within each Sun X4600 chassis the 32 CPU cores inter-communicate with RAM via AMD hyper-transport bus SMP architecture, while Infiniband interconnects between the three nodes, in addition to dual gigabit ethernet connections for external communication.

We will soon be adding a GPU processing capability to the MSG, comprising five nVidia Tesla S1070 quad-GPU arrays.  Some will be added to some MSG-II nodes (for GPU intensive applications) and some to MSG-IV nodes (for hybrid CPU/GPU intensive applications).  This will provide a total of 4800 GPU ALU cores producing between 2Tflop/s (of double-precision) and 20Tflop/s (of single-precision) additional computing capacity at up to 5Gflop/s per watt for those applications that suit adaption for parallel processing on the GPU architecture (cf 1000cores for MSG CPUs delivering 7Tflop/s at 200Mflop/s per watt). Please see this page for further information.

Table 2 - Default MSG memory configuration

Table 3 - Special MSG-IIIe memory configuration options available on request

Each MSG-IIIe node has 40 GB of RAM plus a similar amount of virtual memory swap space allowing MSG-IIIe memory to be configured in a variety of flexible ways while still maintaining minimal practical probability of physical memory contention, i.e. minimal swap-space load. Please contact Philip Chan  to configure the MSG-IIIe job slots to meet specific memory resource requirements of your jobs.

Monash thanks Sun Microsystems for their continuing generous support of the Monash Sun Grid.

The MSG has pools of nodes running:

• Linux: Scientific Linux 5.1
• Sun Solaris v10 and Java
• Microsoft Windows XP

Related projects

Monash Large Data Store (LaRDS) Project
This project provides hundreds of tera bytes of additional storage for Monash researchers.  This storage is available for access via the Monash Sun Grid and other platforms.

Monash Campus Grid (MCG) Programme
Together, MSG and LaRDS form major foundation components in the development of a Monash Campus Grid (MCG) facility.