Cluster Confusion
I attended a briefing put on by Dell's High Performance Computing (HPC) group this week that brought to light a point of confusion that I hadn't really thought about before. The whole area of clustering seems to be greatly misunderstood by the large majority of computer users. I bet if you asked Joe Average computer user about the purpose of a cluster you'd get the response "to make things run faster." Partly true.
Clusters have gotten a fair amount of press in the open source community as a way to group together a large number of inexpensive computers and get lots of computing horsepower as a result. The Beowulf project (http://www.beowulf.org) was one of the first to highlight Linux in a clustered environment. Getting the most out of a Beowulf cluster requires specialized programming to take advantage of all those processors. Meaning you probably won't get any more performance out of your J2EE application unless you do some major recoding.
For many years the idea of a cluster meant high availability / reliability. If one node in a cluster failed a backup node took over. Novell sells NetWare 6.0 with a 2-node cluster license built right in. All you have to do is run it on a system with a shared disk environment and you're good to go. Most of the major server-class hardware vendors will sell you an off-the-shelf system specifically designed to work in a clustered environment.
At the Dell presentation they made a big deal out of the fact that no two clusters are exactly alike. They'll sell you a cluster if you'd like but you won't really get the most out of it unless it's tuned for your specific application. Understanding the problem that you're trying to solve will help in determining where to put your money to get the most bang for the buck. They even have a lab setup in Austin, Texas where you can bring your application in and test it out.
There will come a point in the not too distant future where how we do our jobs will not be bound by the computing resources available. As an engineer I have often had to "find a way around" a particular problem because of the limited disk space or a slow network connection. Pretty soon we'll have to start answering the question "if you had unlimited CPU, disk space, bandwidth, etc., how would that effect the way you do your job?"
7:58:32 PM 
|
