Clustering

[tam2]

I currently have 12 spare pc's with the following specifications:

Intel P4 2.4Ghz
512-1024Ram
80GB HDD


I am looking to cluster these machines, I'm looking to create a Windows or Linux Cluster and would like to know how to do it both ways if possible?

Would there be any benefit of clustering these machines?
What would be a good use for these machines if they are clustered?

- Thanks

[tehboris]

For clustering Windows (as far as I am aware) you need Windows 2003/8 Advanced Server, and each computer would need it's own instalation which m,eans a license per machien (very expencive).

On Linux you would only need to install a OS on one of the computers providing the rest are able to boot from LAN.

There is actualy a live CD for Linux that has all this functionality built right in. It's called ClusterKnoppix (http://clusterknoppix.sw.be/). Unfortunatly this project has fallen by the way side and is using the rather old 2.4 kernel which means alot of new hardware (the biggest concern been network cards) doesn't work with it at all. There arn't many resorces (that I know of) for building a cluster from scratch, sorry on that part.

[latenitetv]

http://www.beowulf.org/

maybe this will help...

[tam2]

Thanks for the replies, ill have a look into them.

[computoman]

if you want a cluster, you must be more specific as to the type you are looking for. Here are two basic ways to cluster.
-
First is a hpc (high performance cluster) where you try to get the most performance from one particular program or programs. It may take special coding in the programs to take advantage of this method.and not just any application software will work. Mostly used where speed in solving a problem is the critical issue. I think Virginia Tech has about 1000 or more macs clustered to do just such a project. It is one of the top clusters in the world.
-
The second is a hac (high availability cluster) where you are looking for dependable and failsafe system that uses redundant equpment. This is where you want a system that will not go down. It might be used for web servers where dependability is critical. Sometimes the work load will also be properly shared between all the machines so that no machine gets more burden than any other machine in the cluster. If one machine fails, the other systems take over the load. This is sometimes known as load balancing. Sometimes mirroring is used also so that two machines are running the exact same program at the same time. Doing mirroring usually requires using special hardware to connect the mirrored machines and the duplicate raid storage.
-
Beowulf is one of the most popular ways to cluster for linux boxes. Cluster knoppix is certainly one live cd. rockscluster and pelicanhpc are others. A cows (cluster of workstations) can be made from almost any cots (commodity (computer) off the shelf) systems. I love the cows terminology being from Texas. That way I can say I am going back to the farm (server farm that is) to take care of my cows. The original clustered machines were made from old i486 class computers. The main trait of a cluster is that all the machines are networked on their own private network. (that may mean adding a second nic (network interface card). Technically any computers networked with their own private network could be called a cluster, I am sure you are looking for something more than that. Where I used to work we originally had nt4 clusters and eventually went on to the win2k series os. The original win2k cluster i think requi\red using scsi drives which made it expensive,. Since I have not kept up with MS. I am not sure what win2k8 or windows7 server will require. Get ready to spend big bucks. Linux is probably the cheapest way to cluster. I have seen people do it with the original xboxes. More and more personal clusters now are being run on the ps3 under yellowdog linux. In fact the people who make yellowdog have come out with a desktop/server version of the ps3. it is not cheap for the average user.

Good luck! Clustering is where the big bucls are being spent now.

[gimpbully]

I think the most pertinent question here is what do you aim to accomplish with a cluster. From a basic standpoint, you can just install any linux, set up passwordless ssh-keys between all the machines, bam, you got a cluster. From there, you'd normally like to install mpich or some mpi implementation, set up a batch queue/scheduler (maui/torque, loadleveler, etc) and get some code running... but if you don't have mpi code... that might be complete overkill..

I currently have 12 spare pc's with the following specifications:

Intel P4 2.4Ghz
512-1024Ram
80GB HDD


I am looking to cluster these machines, I'm looking to create a Windows or Linux Cluster and would like to know how to do it both ways if possible?

Would there be any benefit of clustering these machines?
What would be a good use for these machines if they are clustered?

- Thanks