The Itanium platform gets a boost today as Intel Corp and Japanese server market Fujitsu Ltd announce a strategic alliance to develop "mainframe-class" servers based on Intel's 32-bit Xeon and 64-bit Itanium processors,writes Timothy Prickett Morgan
The enterprise servers borrow heavily from the experience of Fujitsu and its partner and co-marketer, Germany's Siemens AG, in creating the PrimePower line of Sparc-compatible RISC/Unix servers. They will run a beefed-up version of Linux and will eventually scale to 128-way machines by 2005 using the "Montecito" dual-core Itanium chips from Intel.
The Intel-Fujitsu alliance will also create enterprise-class servers using the 32-bit Xeon chips, and while details are sketchy, it looks like they will not scale to these heights, perhaps topping out at 32 processors, maybe 64 processors.
Both Xeon and Itanium machines created by the Intel-Fujitsu alliance will be sold by Fujitsu in Japan, Fujitsu Siemens in Europe, and Fujitsu Technology Solutions in North America. They will initially run Linux, but Jack Hirano, director of communications for Fujitsu America, says that the project will also create Windows variants of the machines.
The future Xeon and Itanium servers will be developed alongside existing Primergy Intel-based machines that are in the works from Fujitsu and Siemens. The exact specifications of the machines are not available, but the largest Xeon machines (probably a 32-way running a kicker to the current "Gallatin" Pentium 4 Xeon MP chip) will ship at the end of 2004 and the 128-way Itanium machine will ship in 2005. Exactly how much technology will be borrowed from the PrimePower line is unclear, but Richard Dracott, group director of enterprise computing at Intel, says that the future product line will span the categories and capabilities of the Sparc-based PrimePower line.
While Intel is obviously happy about the endorsement that this deal gives to its Xeon and Itanium chips for enterprise computing, Fujitsu and Siemens clearly wanted to get something else out of the deal besides enthusiasm, and while financial details were not announced, what seems clear from the limited information is that Intel is rolling up its sleeves and helping out with the substantial amount of hardware and software work that will be required to make Linux scale to 128 processors.
Moreover, in the chess game that is the server business, it will not be surprising to see Microsoft Corp kick in some engineers and/or money to see the project's Windows aspirations reach their targets. With Unisys being essentially the only high-end server maker that has products that scale to 32 processors supporting Windows Datacenter Server (IBM and HP are working on 32-way and 64-way machines), Microsoft needs to foment innovation if its own aspirations in the datacenter to be turned into reality, and then, money.
Building big iron is relatively easy for Fujitsu, and Intel knows how to make chips to plug into servers. A lot of the effort these two will be putting into the future machines is making Linux scale. SGI, which announced a 64-way Linux supercomputer a few weeks ago, put a lot of work into extending Linux and making it work efficiently on its shared memory, parallel machines. While Red Hat Linux will run out of the box on the 64-way SGI Altix 3000 machines, if you want it to run really well, you have to buy all the rejiggered Linux code that SGI has created and which is not being put back into the open source community by SGI. It is hard to believe that Fujitsu and Intel, having done work that differentiates the Fujitsu and Siemens platforms running Linux, are going to put that work back into the Linux community. Dracott says that while no decisions about IP have been made, a model that is similar to SGI's way of doing things is likely. Moreover, it is hard to imagine that IBM, HP, or Dell, if it ever delivers a 32-way or larger server, will behave much differently.
Fujitsu and Siemens' exact plans for supporting Windows on the future Xeon and Itanium machines are even less clear at this point, but Hirano confirmed that it was in the works. Datacenter Server 2003 seems like an obvious candidate for any machine that comes out at the end of 2004 using Xeon processors. What happens on the Itanium machines is dependent on how well or poorly Microsoft and its partners support Itanium.