Intel’s upcoming new chip architecture, code-named Nehalem had a design based on a straightforward principle of modern system design. To run a computer you need a microprocessor and to run a very powerful computer, you need a lot of microprocessors stuck together.
To run them efficiently you need an efficient architecture and, according to Intel, efficiency is what Nehalem, an eight-core Xeon, is all about for the chip is aimed at the datacentre and especially at the largest datacentres or mega datacentres.
Mega datacentres run by companies like Amazon and Google and others of enormous size like that account for around a quarter of all the datacentres in the world. Cloud computing is helping to fuel the growth in the size of datacentres and according to Intel, 20 to 25 percent of all server shipments will go to mega datacentres by 2012.
The issue that Nehalem is supposed to address is how to optimise the datacentres so that they work with maximum efficiency? As datacentres are getting bigger and bigger and are based on more and more microprocessors linked together, that is no small problem.
Intel is working on that crucial issue and at a briefing on Tuesday, Intel’s general manager for high density computing, Jason Waxman, outlined that plan. “We are developing a cloud architecture aimed at mega datacentres with hundreds of thousands of servers that can be balanced automatically,” he said.
The datacentre design will allow for centres that can be balanced and re-sized automatically, Waxman said.
Waxman also said that Intel will use Nehalem to spearhead a full push into mega datacentres. “We've designed a server for a Nehalem-based board that's optimised for our cloud-computing infrastructure,"
During the course of his talk, Waxman was in effect outlining a dream for anyone who is involved in running a datacentre. Provisioning datacentres is a pain for many IT managers so it would be nice if that could happen easily if not automatically and even better if the systems and the chips on it can dynamically shif priorities depending on circumstance. In a large multiprocessor (and almost all computers today are multiprocessor) dynamically re-allocation of resources are good.
The Nehalem processor will be officially launched later this year.