IBM, Intel open blade switch specs
Standards? Don't hold your breath
Server maker IBM and sometime-server maker Intel partnered in 2002 to jointly create and endorse blade servers based on Big Blue's chassis: the BladeCenter. In 2004, the companies opened up the specifications to the chassis so they could help foster a community of blade, switch, and peripheral manufacturers. And today, they're opening up the specs for the switches that go into the BladeCenter designs.
In principle, open specs for a hardware design are always better than closed specs. IBM had hoped that its partnership with Intel would allow it to leapfrog Hewlett-Packard, which by virtue of its Compaq acquisition got the jump on IBM in the blade server racket.
For a while, IBM was ramping up sales and Hewlett-Packard's BladeSystem machines were losing market share because IBM offered products more aligned to enterprise computing. It looked like the partnership with Intel and the opening up of the BladeCenter specs - which were delivered through the Blade.org community created jointly by IBM and Intel - might allow Big Blue to jump way ahead of HP in blades. Or even get its form factors and interconnections established as a standard.
But then HP launched its c-Class blades, which are denser and have lots of good engineering, and now HP is taking back blade market share from IBM. Which is why, in part, IBM wants to shift the topic of conversation away from blade server sales and open specs.
In any event, Intel's Server Systems Infrastructure organization - which makes servers on an OEM basis for a select number of direct customers and which also licenses its designs to third parties so they can slap their labels or those of their customers on them - is announcing that it has a royalty-free license to the switch specifications for the BladeCenter chassis.
The switch specifications describe the physical size of networking equipment used in the blades and how they fit into the chassis as well as how they attach to the chassis for power and management and how they link to the blades. These switch specs will complement the chassis midplane, blade, and peripheral specs that have already been opened up and which are used by some 200 member companies in the Blade.org community - outfits that are making products for or around the BladeCenter design.
By opening up the switch spec, IBM, and Intel are hoping to help foster innovation in switches, much as it has for other peripherals in the machine.
Intel's SSI biz has 50 customers, including original equipment manufacturers (OEMs, plain vanilla box sellers), original design manufacturers (ODMs, who make stuff that other people brand and sell), value added resellers, and software companies who also need to push or buy lots of iron. SII was established in 1998 and has 45 server designs currently under its stewardship.
According to Scott Tease, marketing manager of the BladeCenter line at IBM, IBM still retains control of the specs on the BladeCenter chassis itself and for the management module in the chassis. While Intel and its OEM partners can build to these specifications and provide input to IBM, Tease says that IBM has to be in control of this to maintain compatibility.
The specs allow companies to build blades that plug into the chassis - military contractor Themis has announced a Sparc-based blade that plugs into the BladeCenter - as well as daughter cards, host bus adapters, and so on. Networking gear suppliers have had to license the switch specs to build gear for the BladeCenter thus far, but now they will be available for free through Intel's SSI organization.
Interestingly, no one outside of IBM and Intel has yet built an x64-based blade that plugs into the chassis. HP, Dell, Sun, and Fujitsu-Siemens certainly don't want to help the BladeCenter cause, but it is a bit surprising that tier-two server makers haven't built BladeCenter compatible blades. "No one has done it yet," says Tease. "But it is not out of the question."
Intel was the designer and manufacturer of the first four-socket blade server that IBM delivered for the BladeCenter, and Intel also designed the BladeCenter T chassis, which is tweaked for telecom and service provider customers. IBM created the current BladeCenter H chassis as part of the partnership and also did the more recent two-socket and four-socket blade servers too.
The opening up of the switching specs for the BladeCenter is also important for another reason. If IBM is worried about anything, it is that Intel might anoint another chassis - perhaps HP's c3000 entry and c7000 high-end boxes - as the blade boxes of the future. IBM doesn't want that, and creating a big community that creates stuff for the BladeCenter is part of keeping Intel and its OEM partners interested. Neither HP nor IBM want Intel creating its own blade server designs either, but if the industry did have someone creating the motherboards and chassis for standardized blade servers, it might do a lot of good. Intel and AMD - working together (if you could imagine it) - might be able to create such a standard.
But there is little margin in blade server standards (and more margin in each vendor doing its own thing), so don't expect any real standard emerging any time soon. ®
"sunshiners/hp/lulz/bwhahaha/Scwhartz/IBM/Sunshiners/linuxandwindows/shareprice/lol/sunshiners etc etc etc...."
Common switches a good idea but unlikely.
It would be great if I could take an IBM switch and plug it into an HP chassis, and vice versa, especially as the majority of the blades switche modules are from the same sources (Brocade, CISCO and Nortel) and even share common core components, but I can't see it happening for two reasons.
The first is that these blades are built to a cost, having tracks, chips and connectors to provide connectiviy for x number of blades in a chassis. The problem here is all the vendors put different numbers of blades in their respective chassis, and then have different points at which the blade connects through to the backplane and the backplane connects through to the switch module. If vendor A has a chassis that has the backplane connectors horizontally in the middle for fourteen blades, but the new common design has vertical middle connectors for sixteen blades, then vendor A has to pay out for a redesign of the backplane and then accept the additional cost per switch module for two sets of conenctions they will never use. This is even before we consider that if vendor A's design has a superior feature that can't be carried across to a common design then they are effectively surrendering an advantage that may be a core reason for their sales, for the limited benefit of a commonality feature that would in effect make it easier for IBM to attack them.
The second is that these switch modules interct in different ways with proprietary firmware in the chassis, so a change of module design may also reauire a complete rewrite of your chassis firmware, and again may force a vendor to give up a superior feature for the dubious added value of commonality.
So in short, seeing as the connections onwards from the different switch modules are all common standards (UTP cabling, FC cabling, etc), it seems like the market is fine with propreitary switches. If IBM was the market leader and had some stellar advantage in their designs then I could see a market push for other vendors to agree a common switch module standard, but at present IBM is not the market leader and they do not have any feature I've seen that gave them anything like the driver necessary for the market to start screaming at HP, Dell and FSC to comply. The only likely taker I can see is Sun as they would actually gain from handing over a chunk of design control to IBM given IBM's superior market position and Sun's linited engineering ability.
Compaq acquisition causes HP blade lead?
> IBM had hoped that its partnership with Intel would allow it to leapfrog Hewlett-Packard,
> which by virtue of its Compaq acquisition got the jump on IBM in the blade server racket.
mhh... HP's lead in blade market share reflects HP's overall lead in x86 servers, which HP inherited from Compaq. If that is what is meant here then it makes sense. But IBM led the blade market until well after Compaq was acquired by HP, so unless qualified the statement above is simply wrong.