Yesterday I was an all day Architect Council Meeting about Grid. We had the Head of Microsoft Research in the UK, Two professors from Southampton University and an Architect from a major Bank who had implemented a huge grid application. I have been looking forward to this discussion for some time as some 15 years ago I was working in parallel processing research at IBM so was interested in how grid overlapped with parallel, how far Parallel Processing has come in 15 years and what all the hype on Grid was about.

 

Well it certainly was an interesting discussion. It seems that there are two elements to the grid story; the applications that can be run on the Grid and the platform that they run on. This then maps pretty well into the reality and the marketing as well. The reality is that most if not all successful grid applications out there are the parallel processing applications of old, Monte Carlo, CFD, FEA, TSP; the heavy duty scientific algorithms that have been around for a long while. What the marketing is talking about as grid is general purpose business applications running on a grid platform with the same distribution, scaling and performance characteristics as the scientific algorithms. The problem with this story is that it is not true. The Grid applications today have a huge amount of homogenous concurrency which can be very easily and simply distributed which is simply not true for the vast majority of business applications.

 

As with all marketing however there is a grain of substance in the claims. It is possible to use a grid platform to write distributed business applications such as we have been writing for the last 15 years. Of course the mechanisms, architectures, tools, issues, performance, scalability and support of these applications are exactly the same as in the more conventional distributed application world.

 

So the Grid marketing story seems to go: Grid Applications work efficiently on a Grid platform (true) so therefore general purpose business applications will work efficiently on a grid platform (Untrue). Thus we just have to build a great grid platform (or at least story) and business applications will become instantly and easily scalable. This would be laughable if it wasn’t for the fact that people actually believe it.

 

The really sad thing about this deceit is that it will sully a couple of very important things that the grid community are bringing to computing. These are:

 

  • There are a very important but distinct set of computation or data intensive tasks which grid works extremely well for.

 

  • The concept of a grid platform or system to run distributed applications on is a powerful model which has very broad applicability.

 

Unfortunately most of the Grid effort seems to be going into messaging systems which has already been done (to death!) in the more mainstream computing platforms with the Web Services Standards. Where Grid has done some innovative work has been in the less sexy but more important areas of management and operations. If this experience and expertise was taken by the grid community into the mainstream platform as standards and products then they could have immense value and influence.

 

So to answer the questions I had at the beginning: Successful Grid applications are the parallel applications of old, parallel processing hasn’t changed much in 15 years and the Grid hype is just that; hype.