Big Blue is fired up about the potential of the blade server market, not because of the number of orders it has taken – Hewlett Packard entered the market a year ago and has shipped 15,000 blades to date – but because of the level of excitement that a relatively thin number of offerings among all the players have generated among the enterprise customers who are IBM’s bread and butter. While many enterprises are still testing and prototyping blade machines, the prospects of consolidating machines into more dense and easier to manage collectives has opened up CIOs and MIS managers to the possibilities of buying blade gear as a means of transforming at least some of their IT operations.
Jeff Benck, director of eServer product offerings at IBM and the guy who sings the praises of the BladeCenter for Big Blue, says that the usual gang of industry analysts have underestimated the potential of the blade server market, and the fact that IBM and HP can ship so many blades in such a short time so early in the bell curve of this new technology might suggest that this market could take off a lot more than many people think. Sun Microsystems Inc has just jumped in the game this week with its Sun Fire Blade Platform, a few months after Dell Computer Corp came out with its PowerEdge 1650MC modular servers. HP has uniprocessor, two-way, and four-way blade servers, and is keen on further fleshing out its offerings in this space to maintain its market share leadership. IBM only has two-way blades at the moment, but Benck said that IBM was working on future blades that would plug into the current chassis.
The BladeCenter is a 7U form-factor chassis that can house up to 14 two-way server blades, yielding a total of 168 processors in a standard 42U rack. The chassis has an internal Gigabit Ethernet backplane that the blade plugs into, and also includes Ethernet switches and, in the future, will have Fibre Channel and InfiniBand switches as options. The BladeCenter H20 blade, which plugs into the chassis vertically, is based on the ServerWorks Grand Champion-LE chipset and it can have one or two Prestonia Pentium 4 Xeon DP processors, which are equipped with 512KB of integrated L2 cache memory and which run at either 2GHz or 2.2GHz. The HS20 blades supports from 256MB to 4GB of main memory.
Benck said that IBM was readying a four-way blade server for the BladeCenter chassis for later this year, which is being developed by IBM and Intel Corp as part of a blade server alliance the two formed last year. This four-way blade will be based on the Gallatin Pentium 4 Xeon MP processor. While it will plug into the same chassis, it will occupy two blade slots, which means the processor density of the four-way blades per rack will be no different than with two-way blades. The difference, of course, is how much processing can be brought to bear per blade. Database and application servers sometimes require four-way processing, even in clustered environments.
While HP and Sun offer uniprocessor blades, Benck says that IBM has looked into the market for this and has come to the conclusion that it is not necessary. He says that if customers want to, they can simply half-populate the H20 blades. But the density per rack is then a mere 84 processors, which does not compare to the 280 processors or 224 processors that the HP and Sun designs can bring to bear for light infrastructure workloads. It’s about the same density as you can get using regular two-way, 1U servers, in fact.
He confirmed that the two-way PowerPC blade that IBM was demonstrating at trade shows last fall was indeed going to use the future PowerPC 970 variant of the Power4 processor. The PowerPC 970 is a stripped down, single core implementation of the dual-core Power4 processor that is widely expected to be the volume processor used by Apple Computer in future Mac desktop and Xserve server machines. Sources say that IBM will debut the PowerPC 970 in a two-way blade running at 1.8 GHz. This chip will also, we are told, end up in entry pSeries Unix workstations and servers. Like the Power4 processors, it can support AIX or Linux, and apparently also has all of the circuitry to even support IBM’s proprietary OS/400 operating system.
Benck also put a little color on IBM’s Itanium blade plans. Last year, IBM was hinting that it would offer an Itanium-2 blade, but wouldn’t say if it would be using the current McKinley generation or the future Madison generation, which is due around mid-2003 if Intel can keep to its schedules. Contrary to some of the buzz going on in the industry, IBM most definitely has Itanium servers on its roadmap, and blade servers based on Itanium as well. Benck says that the current plan is to use a McKinley two-way blade that takes up two physical slots in the BladeCenter chassis. This is necessary because of the heat that Itanium throws off. Itanium is, for certain HPC applications, something that some customers will want, particularly for companies who want to create dense Linux clusters. Of course, IBM will want to peddle its PowerPC 970 blades in here as well, which will also support Linux.
Finally, Benck made a comment that suggests that IBM was exploring the possibility of adapting its Summit chipsets for IA-32 and IA-64 processors to create hybrid blade-SMP/NUMA designs that would allow companies to create what is effectively an SMP server across multiple blades in the chassis. The current xSeries 440 design uses what IBM calls scalability ports to turn four four-way xSeries servers using the Summit chipset into a 16-way machine with a single system image. Summit can scale beyond 16-way clustering, and could be used to gang up uniprocessor, two-way, or four-way blade servers. Such a development would allow companies to effectively create SMP servers on the fly as they needed them from a common core blade component. It is an interesting idea.
Source: Computerwire
