How SSI Cluster differ from PVM Cluster?

SSI:

SSI stands for Single System Image which is the simplest way of High Performance Computing, it uses a clustering middleware such as openMosix (http://openmosix.sourceforge.net/). In the SSI cluster, all the member nodes in a Local Area Network collectively deliver performance as single virtual machine.

Advantage:

• Very easy to setup.
• No special hardware and software requirements.
• Applications need not to be recompiled for multicomputing environment.
  

Usage:

Batch processing aaplications such as Backup(tar+zip) for mulitple files, Media Format Conversion for mulitiple files, Content filtering and e-mail virus scanning etc.

PVM:

PVM stands for Parallel Virtual Machine, in which the cluster distributes a single task into crunches and processes them individually over the member nodes of the cluster. However, a major issue with this type of custering is the recompilation of the application with the PVM/MPI (Parallel Virtual Machine/Massge Passing Interface) libraries such as OpenMPI (http://www.open-mpi.org/) or OSCAR (http://oscar.openclustergroup.org/).

Advantage:

• Can distribute a single task requireing higher CPU Throughput into multiple sub-tasks.
• Node job assignment and thread management is taken care by parallely compiled application itself.

Usage:

Applications that require more CPU crunching such as scientific applications, data modelling, weather forecasting etc.

Three Extreme Distors in a Month!

The April month of this year proved to be the a milestone in releases of various popular GNU/Linux distros. It started with the release of longly awaited Debian 4.0 (aka Debian etch) http://www.debian.org/releases/stable/ on April 8, 2007. Then there was the turn of CentOS 5.0 http://www.centos.org/ and finally Ubuntu 7.04 (aka Feisty Fawn) http://www.ubuntu.com/ was released on April 19, 2007.

Benchmarking Cluster Performance with Linpack 1.0a

Everyone knows that Linpack (http://www.netlib.org/linpack/) is the de-facto standard for benchmarking a multiecomputer's performance under load conditions caused by complex mathematical computations. Even the official high performance measuring project Top500 (http://www.top500.org) also uses Linpack itself as a yardstick of high performance computing measurement.

I also tested and benchmarked the performance of a small cluster through the High Performance Linpack benchmark 1.0a (http://icl.cs.utk.edu/hpl and http://www.netlib.org/benchmark/hpl/) and found the following results.

In addition I must comment that it took no time to set up a cluster by using the ParallelKnoppix distro (http://parallelknoppix.cebacad.net) as the distro itself being a live-cd based on auto node discovery.

So, let's talk lesser and see how were the results??!!

-----------------------------------------------------------------

Benchmarking Date: April 13, 2007

Platform: ParallelKnoppix version 2.4 (23-Feb-2007)

No. of nodes:3 + 1 Master machine

CPU: 1 x Pentium 4 HT @ 2.4 GHz on each node as well as server

RAM: 256 MB on each node as well as server

Inerconnect: Gigabit Realtek NIC connected through 24 port 100 mbps, HCL fast ethernet managed swtich

-------------------------------------------------------------------

And the benchmarked result was 2.35867 Giga-Flops!