how do you construct such a cluster. what connects the computers ethernet, can wireless do it? what software is needed?

A wireless connection will work for Beowulf clusters, but the limited bandwidth of wireless networks(~11mbps) severly limits the through put of the parallel algorithms. They need to communicate large amounts of data very quickly to the master node.

Beowulf clusters are hetergeneous, meaning they can be alpha, intel, amd, risc processors. Just about anything will work together. Different processor speeds are also allowed on the same cluster. There are two "mainstream" controllers of Beowulf clusters: Parallel Virtual Machine (PVM) and Message Passage Interface (MPI).

To construct such a cluster you simply build as many machines as you can, multiples of 2 make life easy. My cluster is currently 4 machines. The connection between them is very important. I have a 10/100 ethernet connection.

The problem will be writing parallel code. There are several programs out there for you to use. You have to ask yourself what you want it for. If it's for fun, then search the net for parallel code to rip mp3's or to do some bazaar computation in a short time. Also, POVRAY can produce nice CGI's.

Jon

i mainly want to use it to help speed up the kernel recompile and software installation on my slower computers

can anyone tell me how to connect the computers and what if any software is needed? thanks.

That's my question as well, but another is does this work with Windows machines? (2000 and higher)

But how are you going to parallelise the code? The kernel is written in C, and has support for multiprocessing but multiprocessing is not the same as parallelisation which is the main application for cluster computers.

Most problems actually can't be parallelised. Think of this: if a woman can produce one baby in 9 months, can 2 women produce 1 baby in 4.5 months?

A smarter solution would be to download one of those distributed application SDKs, and to ask all your friends to run it on their desktops.

so, is there anyway i can hook up my 800mhz desktop, to my 233mhz laptop and have it do compiling and recompiling of software and kernels.

1. if a prog isnt multithreaded can i have it just use the desktop cpu
2. are multithreaded apps the only ones that can be beowolfed?
3. how do i set up a beowolf cluster? how could i take full advantage of it on my slow laptop?

A multithreaded application can use a single CPU too, but it won't execute it in parallel. Parallelisation and multithreading are not really the same - many programs have more than one thread of execution - the OS you're using now certainly has more than one thread of execution. Think of the most visible application of parallelisation (and distributed computing), SETI. This task can broken into many small independent pieces and then reassembled to give a result.
Parallelised programs are the only ones which will take advantage of the architecture.
If you only want to save time (2 babies in 4.5 months), think of NFS so that you can link up the two computers over TCP/IP and run the kernel recompilation concurrently. Careful though! If I'm not mistaken, kernel recompilation is a sequential process. You want to use two processors to share the workload? Sell one of the computers and use the cash to buy more RAM and a faster processor for the other one. Or get out your soldering iron and put the second processor into the other computer if thats possible on your hardware.

Sorry to shake yours ideas of clustering and beowulf etc., but there's a lot of myths about this topic - it's not for the light-hearted, and would not really pay off for less than about 20 computers - standalone PCs are just not put together with clustering in mind - they are too autonomous in the way they process instructions.

Bert

ok, thanks for the bad news.

I am currently about to take on a project at school to create a beowolf cluster as a senior project, not quite sure how to do this as in the software end. Most like going to be a couple dozen 486's with a 10/100 connection through each in a custom box, what software should I use and how to I use it

Try this site:

http://dune.mcs.kent.edu/~farrell/equip/beowolf/

They have an FAQ there that kind of explains the challenges. Most of the other links appear to be dead.

However, here is a link to a how-to

http://tldp.org/HOWTO/Beowulf-HOWTO.html

I did a beowulf cluster on 35 PCs running RedHat 7.1 as OS for my end-of semester thesis..

well, software side, there are quite a lot out there, but the one i used are MPICH (for MPI) and PVM...MPI is quite new (at least here in my country ) and PVM is more mature and more widely used here. I wrote parallel programs (mathematical problems) using C++ to run on the cluster and compared the two (MPI and PVM), studied the load-balancing and stuff..

actually it wasnt quite that hard..u just have to configure the RedHat box a little, most of it for cluster administration..and RedHat 7x even comes with this sofware called LAM-MPI, which is similar to MPICH, but i didnt try it coz some people said it's not as sturdy as MPICH or PVM..

the beowulf link rkaran gave is a good one...if u guys really interested i'll refer back to my tesis and see if i can translate the whole 250 pages of it into English or something hehe...

but really, this cluster stuff doesnt really do real work(at least in my case, parallel programs/computing) unless the cluster consists of more than 32 PCs at least, but even then,using 35 PCs, i'm not much impressed...

Suzana

I have docs, RPMs and all the tools needed for clustering. The HOWTO I have is a step by step cookbook-recipe type of thing.

Email me at chupacabra@linuxmail.org.

Do you have a place where I can upload that stuff? total weight is about 7MB.

email it to me, i can try to put it on my crappy web site... which is crappy... http://home.earthlink.net/~goatdemon/data/

my email is goatdemon@earthlink.net

Check
www.rocksclusters.org
They can help ;-)