[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: kernel config used for cluster knoppix

In article <buh5sf$mpv$07$1@news.t-online.com>, kegwasher wrote:
> Thanks for the hint I will check there.  I have used the knoppix cluster CD
> to make a 4 machine cluster with two waiting on parts.  

> The knoppix/debian
> packages have MPI available but I am trying to figure out if the Syclad
> beowulf version is better or just more hype.  

It depends on what your criteria are. It is not free, but it gives you
the best method of syncronizing the installations, and you will be increasing
your performance and boot process:

- the software is only installed on the master node
- the clients can boot from the master, and they can be diskless and boot 
  via PXE over the ethernet, or if your hardware does not support that, 
  with a small bootloader on the harddisk or bootfloppy. 

So you avoid version skew - if a client runs old software, just reboot it :-)

Performancewise: Only a very minimal kernel runs on the compute nodes, and
only two low-impact processes: one process accepting new jobs and one process 
returning status info to the master. 

All the rest of the CPU/memory is dedicated to your actual compute tasks.

And you can just plug in another node into your network, and have it join
the cluster simply by PXE booting it. No installation necessary.

> Looking to do some CFD and
> FEA analysis and want to maximize my output while minimizing my input.  I
> am also trying to work out the best method of syncronizing the
> installations without hurting performance or making the boot process take
> to long.  My current thought is to use a list of .debs on the server and
> make apt-get look there from each node at boot to ensure all are playing
> the same song on the same page.

Michael Will
Michael Will, Linux Sales Engineer
Tel:  415.358.2673  Toll Free:  888-PENGUIN
PENGUIN COMPUTING - The World's Most Reliable Linux System

Reply to: