waht does it take to set up a cluster

watts3000

Senior member
Aug 8, 2001
619
0
0
I would like to know when setting up a ms cluster do the servers have to be exact hardware? I've read that they must match but I want to know how closely must they match. Also what do you guys use for shared storage when setting up a cluster.
 

drag

Elite Member
Jul 4, 2002
8,708
0
0
I don't know what it all takes hardware-wise for windows stuff, but you need to define exactly what "clustering" is.

You have clusters that are used for massively parrallel proccessing. Were your doing major number crunching by dividing up the workload between numerous machines.

Then you have clusters were you provide a seperate workload, say a file server for a large network. Instead of 1 big server you have 2 small servers. You can also mirror servers, were you have 3 servers, were 2 are working and 1 is standing by incase one of the other breaks. If one breaks then the stand-by instantly becomes active and replaces the functionality of it.


Windows can only do the latter, not the former. Unix-based OSes like Linux (especially linux, but OS X and the other BSDS can do it, too) can do both types. I realy don't know any large scale windows clusters being used anywere, but a major example of Linux being used is Google.com. They use both types of clusters were you have nodes of several machines that take and divide of one workload. Then they have several nodes doing different stuff and a couple nodes just waiting to be backups. Hundreds of machines working together to provide your search engine needs, all admin'd by a small group of people. Another big example is the "Big Mac" OS X cluster used for massively parrellel proccessing. It's like the 3rd most powerfull computer in the world right now. Numerous Linux machines litter the top 500 computer list, too.

The most well known is the Beowolf cluster. This cluster you need to write special programs that can utilize the cluster, and is used in scientific-style calculations. (protien creation, mapping star movements, nuclear research, etc etc). A easier Linux clustering method is OpenMosix. This can improve the performance of regular programs. This actively moves programs proccessing from one computer to another to divide up the work load. No single program (unless it's multithreading like some webservers or 3-d animation rendering) will ever go faster then on the fastest single proccessor, but if you have several things going at once then you can get a big boost.


If you want more information on windows stuff you can check out pages like this were they are discussing parrellel proccessing of stuff or microsoft.com were they have some techinical doc pages, but right now Linux is light years ahead of windows in this sort of thing.

The easiest (vastly easiest and cheapest) way to set up a cluster to play with would be to check out OpenMosix's webpage. You can then set up one box to be the cluster server, and then get something like ClusterKnoppix or Plump linux or Gentoo Open Mosix live CD.

THose will provide ways to set up diskless clients from just a CDROm or network boot that will make it much easier to get a cluster network going. The good thing is that you don't need identicle hardware, Open Mosix will migrate the programs based on the network ability and performace/memory aviable from each machine. Kinda try to balance everything out.

Unfortunately my understanding of windows clustering technology is pretty limited.