Linux Cluster Help?

[Jon855]

I'll give you a general idea of what I'm trying to do. I have a node I would like to be booted by the network in which will be connected to switch and the server will be connected on the same switch. How do I use the other computer's CPU in assisting the main node "server" doing its job or is it done differently? I run RH fedora Core 3

 

[Nohr]

I don't know much about this myself, but try looking up info on beowulf clusters.

Engineering a Beowulf-style Compute Cluster (LONG read)

 

[drag]

Check out:
www.openmosix.com


That will help you setup your system so that you can migrate proccesses from computer to computer. It's like a SMP machine were it won't make any single job run faster, but if you can split it up in parts you can move it to other machines to proccess it faster.

They have bootable cd's and such that can help you set it up. This is probably the easiest way of doing it.

Beowolf clusters are designed for specific applications and require that you program them to use special message passing libraries and all that. It requires applications specificly programmed to be used on such clusters. OpenMosix on the other hand works with mostly any application.

One example is that is used is when compiling code. When using 'make' to compile a kernel or whatnot you can tell make to spawn several instances of gcc compilers and those can be migrated to other nodes.

Another thing along OpenMosix lines is OpenSSI. http://openssi.org/cgi-bin/view?page=openssi.html

It will migrate proccesses similar to openMosix but it's a bit more comprehensive in terms of setting up a 'single system image'. With it it will help you setup your machines as one big machine so that you can share a network address across several machinesm, have one shared root filing system across many machines, and do stuff like that. It won't turn your system into a Beowolf cluster, but it can be used to help you setup something like that. It uses OpenMosix algorythms and such when it comes to moving individual proccesses from machine to machine.

Generally what you do is have a master node, then the other machines boot off of the network and download the kernels over the network. Then then system gets root and is setup like a regular machine.

You can do stuff like run commands on different machines by going:
onnode 2 dosomething
onnode 3 ls

and stuff like that.

OpenMosix will be the easiest, and I think it's a bit more mature.

 

[Nothinman]

How do I use the other computer's CPU in assisting the main node "server" doing its job or is it done differently?

It's not that simple. You need tools to distribute the load or programs designed to distribute their own load. You can't just run a process and have it use both CPUs as if they were 1.