Clustering in Linux.

[trinak96]

I'll admit to being a "newbie" but i'm interested in the possibilty of creating a clustered computing environment using Linux.

In theory i want to create a collection of older (spare bits!) pc's into a single, useable entity for any cpu/memory heavy application.

So, in practice, i would have a single desktop environment with applications installed, and be able to run any of these apps, which would utilise the memory and cpu of all devices making up the "virtual PC".

Questions:

First : Is this possible? If "yes" then please refer to next question, if "no", so be it.
Second : How ?

Many thanks in advance.....

[Juan Pablo]

Yes, it's possible. OpenMosix is what you want but it's not easy in any way. Take a look here openMosix, an Open Source Linux Cluster Project

[trinak96]

Thanks Jaun Pablo,

I have had a breif read through, that's the sort of thing i'm after but for one problem. It seems that a single process will not be shared across the cluster.I'm looking to be able to have a single process running across the entire cluster. Is still a possibility or are we getting into the realms of a re-write of the original code in order to use all resources?

Thanks again...

[vsemaska]

What you're thinking of is not the way clusters usually work. A process runs on one node in the cluster. The advantage of a cluster is if that node fails the process can be restarted on another node. So one of the things clusters provide is high availability.

To have a process running on multiple nodes it would have to be multi-threaded and the threads would be running on different nodes. A rather difficult thing to do plus you get into issues like the speed of the interconnect between the nodes. Now your getting into the realm of very high end and very expensive cluster systems.