Grids, on the other hand, are great at carving up some process -- examining signal data in the search for extraterrestrial life for example -- into discrete chunks and then parceling out those chunks to distributed systems (often dirt cheap ones) that digest them and return the results for re-aggregation into an answer to some problem. The more systems in a grid (no commonalities required), the more chunks that can be parceled out simultaneously, and the faster you get your answer.
But what if you wanted the best of both worlds: The ability to distribute tasks to dissimilar commodity hardware while simultaneously hardening the overall “system” to be so fault tolerant that it practically never goes down, without breaking the bank? Sounds like an impossible dream, doesn’t it? According to Appistry's Sam Charrington, you can have your cluster cake while eating the grid too.
The premise of Appistry's Enterprise Application Fabric (EAF) is simple. Instead of going overboard with a high-end farm of Xeon servers, go the other direction with a boatload of dirt cheap Pentium 4s. With EAF, you insert some wrappers into your application's code and the result is a hive of computers that operates the same way a beehive does. If a worker bee gets killed, the other computers just absorb the workload. If a new worker bee is born (aka: a new computer is added), it is automatically assimilated into hive and given a job to do. One of Appistry's premises is that when a computer fails, you just throw it away. While you couldn't do this with a high-end Xeon box, you could easily do it with a cheap Pentium 4 box. In some ways, Appistry may have undone all the hard work that Intel and other chip manufacturers have poured into making their server chips mission critical.
In an interview from Gartner Symposium/ITxpo (available as an MP3 that can be downloaded or, if you’re already subscribed to ZDNet’s IT Matters series of audio podcasts, it will show up on your system or MP3 player automatically. See ZDNet’s podcasts: How to tune in), Charrington summed up the value proposition of Appistry's Enterprise Application Fabric with the following statements:
The more machines you have in an application fabric, the greater your ability to scale. We have an application fabric in one of our labs where we're able to run a simulation of the type of transactions that a MasterCard or a Visa might run and in fact our founders came from the credit card processing world so they know a lot about these types of transactions. We're able to do on, a set a machines that costs roughly $70,000 the transactional load of a MasterCard at the peak of the Christmas shopping season...There were about 100 Intel Pentium 4 machines.... not even Xeon servers. The beauty of what we bring to the table is that customers are no longer locked into an infrastructure-centric or hardware-centric paradigm and they can choose the hardware with much broader lattitude than they were previously able to do.
Later in the interview, Charrington described how one of Appistry's telecommunications customers incorporated $288 systems from Wal-Mart into their fabric without a hitch. This is the second time I've heard Appistry's pitch (the first was at Globus World when the company was operating under its old name Tsunami Research) and I'm in as much disbelief this time as I was the last. Either these guys are selling snake oil, or Appistry's solution will be one of the most disruptive technologies to hit the server market in more than a decade.