Hello Frank,
Actually the Wolf Mountian research eventually created a set of technologies exploring the area of true clustering on commodity PC hardware. There was a demonstration of these technologies at Novell's Brainshare developer conference over a year ago. All of this was done as standard research to be able to see implementation of technology for further investigation in how they could be productized. Soon after the demonstration was when the situation developed with a couple of the engineers on the team. If you are interested in where these technologies are starting to show up you can go to novell.com on the Novell site.
The following is a generic description of "clustering" and is not intended to be representitive of the specific Wolf Mountain project.
Clustering is more than disk sharing, it is also CPU sharing. A really good book is "In Search of Clusters" by a very impressive scientist from IBM (I'm drawing a blank on his name right now).
With true clustering, you would have the ability to connect many multi-processor machines (chassis) with a high-speed "fabric" network, and also connect each of these machines to the network. To the users, this cluster appears as a "single system image" meaning that it looks like one computer to the users. In reality the users, and services, and application services are distributed over the various chassis and if one chassis fails, the rest of the machines take over that load and continue operation. These systems will often offer transparent disk access also ... the ability for disks to be "mounted" by any chassis in the cluster. This is usually controlled by the way disks are connected to the chassis - shared access or replicated data.
The problem, as usual, is that Microsoft has "polluted" the word clustering with their Wolf Pack project. (Actually they polluted the word Pack, since I don't think that two wolves make a pack!) What Microsoft created with Wolf Pack is not even equivilent NetWare SFT-III which Novell has been shipping for years. I think that even the Standby Server solution (on the web page) that we are selling today gives more capability than the Microsoft attempt.
In true clusters there is also a difference between "cluster-aware" software and standard software. Much research has gone into this area. To make the most of the cluster, most applications and services have to be rewritten to be "cluster-aware". This means rewriting the software so that it makes cluster-specific API calls to be aware of how many chassis are running the service or app. It is difficult to make the cluster transparent to most existing software. Because of this, most of today's solutions are not "true" clusters, but are "high availability" solutions. This means that the service runs on only one chassis in the cluster, and if that chassis fails it is re-invoked on another chassis.
Sorry ... more detail than most people probably wanted ... but it is a facinating area of technology! We as an industry are probably getting closer to true clustering on commodity hardware, but not quite there ...
Scott C. Lemon |
