|
|
|
Supercomputing has proven to be a boon to researchers across the
broad spectrum of scientific endeavor, drastically reducing
computational time and greatly expanding the envelope of what is
possible in the field of simulation. In the last couple of months,
we have looked at many issues related, one way or another, to the
growing pervasiveness of these monstrous calculating machines.
But what is a small research community, such as a single college
or university, if it just doesn't have the funds to cough up
for a multi-million-dollar supercomputer? One possible solution
to this problem, the clustering of workstations, is gaining a lot
of attention, both from academic and industrial sources. In this
article, we'll look at some of the general ideas involved in such
a scheme and discuss why this might be a viable idea.
To begin to understand how a cluster of workstations might work, it's instructive to think back to the architecture of a generic supercomputer. Basically, such a machine consists of a number of "nodes," each containing one or more processors and some memory element. In addition, there must be some form of communication between the nodes so each can "know" what the others are doing. Now, if we consider a group of workstations, say some PCs with Pentium 350 MHz processors, it's not too hard to pick out quite a few similarities to the supercomputer. Disregarding the input/output devices like monitors and keyboards, each computer has a processor and some memory, so the basic pieces are there. The obvious missing piece is the communication among the processors. Strictly speaking, there generally is some form of communication possible between machines, as they will most likely be hooked up to the net. For communication to be viable, though, there must be some sort of dedicated lines connecting the boxes. While we won't go further into the discussion of communication among the CPU's right now, there do appear to be some ways to accomplish the desired goal here. So, now we have these workstations sitting around that can be connected to each other, but why would we want to do such a thing rather than just using a supercomputer? Well, first of all is the issue of cost, alluded to earlier. A cluster of PC's is going to be a whole lot less expensive than one large machine with the same number of processors. Another advantage for groups with existing machines is that they can squeeze every last bit of computing power out of their resources. At most colleges and universities, there are clusters of PCs in computer labs at different locations around campus. While Go To Page: 1 2
The copyright of the article Clustered Workstations : Block-by-Block Supercomputers in Scientific Computing is owned by . Permission to republish Clustered Workstations : Block-by-Block Supercomputers in print or online must be granted by the author in writing.
|
|
|
|
|
|
|
|