November 3rd, 2004, 01:18 PM
Clustering Windows XP
I am looking for a way to cluster a group of Windows XP machines. Is there a cheap/easy way to do this? They are running Windows XP Pro, they are identical Dell machines, and I have Admin access. There are close to 30 machines. I am just learning about this supercomputing deal. I followed Neutron's ClusterKnoppix tutorial, but found myself stuck as the Knoppix Boot hangs while trying to find the CD in \dev\scd0. I think I need to burn the CD at about 4x.
Anyways, any help with Windows XP Clustering would be appreciated.
Geek isn't just a four-letter word; it's a six-figure income.
November 3rd, 2004, 04:27 PM
XP is a workstation so a "real" cluster isn't possible, you'll need W2k server for that.
You can cluster an application if that application is build for it. But this requires a specific application that's able to decide which machine does what.
Experience is something you don't get until just after you need it.
November 3rd, 2004, 04:29 PM
Their no application that run on XP that can connect a Linux Cluster?
November 3rd, 2004, 05:18 PM
"Clustering" is mainly used in the context of load balancing, fault tolerance and
parallel processing. I think, you are talking about using these computers in the context
of "parallel processing", which is defined as the simultaneous use of more than one
CPU for one executable. Hence, SirDice is completely right with his statement. I
want to elaborate it, trying to answer SDK's question (if I understood it correctly).
If you want to use several machines (XP, 200x, Linux, ...: eg Beowulf Cluster) for
parallel processing you need
a) to write or to have written software that supports parallelisation (using some standard)
b) an interface, which can take care of the communication between the nodes (see also ).
By installing for example MPI[CH] on each of your machines, you are able to send
messages from node to node - messages, which may contain results of the
"threads" and/or are needed for the parallelisation method involved.
If you develop your own code, you have to
(i) be able to parallelise your code
(ii) use MPI[CH] for the communication (quite simple c-functions), watch for deadlocks
(iii) compile your code with mpicc (instead of gcc)
(iv) run your code with mpirun.
Sideremark: There is a branch in the development of gcc 4.0 which includes the support
of OpenMP, allowing you to parallelise by setting pragmas. If you wait another 2 years or
so, parallelised code can be much easier generated.
If the only tool you have is a hammer, you tend to see every problem as a nail.
(Abraham Maslow, Psychologist, 1908-70)