ATCA/Standard Servers

HP Ditches ATCA

HP Inc. (NYSE: HPQ) is dropping its AdvancedTCA products in favor of its own line of telecom blade servers, matching a tactic IBM Corp. (NYSE: IBM) uses with its BladeCenter line.

HP says it's extending its BladeSystem c-Class line to include carrier-class features, a move that replaces the ATCA plans the company announced in 2004. (See HP Extends BladeSystem and HP Plans ATCA Telecom Blade.)

"This is meant to chase the standalone server market where we were selling ATCA," says Christine Martino, VP of telecom platforms for HP's Business Critical Systems unit.

HP lost interest in ATCA as it became "more of a component market" where customers are meant to mix-and-match vendors' pieces, Martino says. HP would prefer to sell complete systems; moreover, it isn't so interested in non-server gear, such as ATCA-compliant media gateways, she said.

HP's move reflects ATCA's state of flux as the industry has reduced the inflated expectations of a few years ago. (See Whatever Happened to ATCA?)

Some notable players have been falling back to their more traditional businesses after deciding ATCA isn't exactly for them, says Ernie Bergstrom, an analyst with Crystal Cube Consulting .

"ATCA never was prevalent in their product lineup," Bergstrom says of HP. "If you went to their Website and looked around, it was never easy to find."

"The big surprise to me years ago was that HP started promoting ATCA for carrier-grade systems rather than competing directly with IBM's BlaceCenter-T," says Simon Stanley, Heavy Reading analyst at large. That HP is now choosing to build off its BladeSystem franchise "seems to be the best commercial strategy" for the company, he says.

Not everyone is turning their backs on ATCA. Intel Corp. (Nasdaq: INTC), for instance, sold its modular communications platforms -- its ATCA business -- to Radisys Corp. (Nasdaq: RSYS) but is pledging to produce microprocessors more suitable to ATCA's tight power budgets, Bergstrom says. (See RadiSys, Intel Deal.)

With the BladeSystem c-Class, which starts shipping in January, HP is trying hard to show it's not just an enterprise-level vendor. The boxes will be NEBS Level 3 compliant, and HP is promising each model will be available for at least three years, with a five-year support commitment after that span. That's eight years during which "the system will be available without any changes," Martino says.

That last point is crucial, because enterprise servers tend to get swapped out every three years. IT departments are accustomed to that, but telcos want something longer-lasting.

HP isn't discussing specific prices but says the c-Class will sell for the same amount as comparable enterprise-class BladeSystem gear.

As with IBM's BladeCenter, HP's approach could have some risks. Specifically, it's a single-company product line competing with the standards behind ATCA, and some customers might balk at that.

"That could happen," admits Martino. "We believe, based on conversations we've had so far, that the majority of them are ready to take that leap."

— Craig Matsumoto, West Coast Editor, Light Reading

Pete Baldwin 12/5/2012 | 2:57:48 PM
re: HP Ditches ATCA A couple of other people had remarked recently about how hard it was to find ATCA on HP's site.

So now we have BladeSystem, BladeCenter, and ATCA. I haven't followed HP closely in recent years, but obviously they're a name to be reckoned with.
Pete Baldwin 12/5/2012 | 2:57:47 PM
re: HP Ditches ATCA Good point.

It's reminiscent of network processors, when Intel started amassing armies of programmers to help out IXP customers.
Hanover_Fist 12/5/2012 | 2:57:47 PM
re: HP Ditches ATCA One point commonly overlooked (or conveniently not addressed) by most ATCA manufacturers/vendors is that complexity in software kills their ability to provide COMPLETE, ready-to-implement, time-to-market solutions.

They would fail to address the biggest issue (or better yet, dance around with slick marketing slides) that using Common off-the-shelf HARDWARE components does not result in a faster time-to-market solutions because they ignored (e.g., avoided) the fact that the software which made the whole system work was in-fact a customized per system work endeavor far more complex than the ATCA hardware specification.

Bending sheet metal is a simple task compared to the millions of programming man hours required just to finish off ONE ATCA solution.

Ask each of the vendors mentioned in this article about the size of their ATCA software programming staff - I'm going to guess that you will be surprised at their answer.

Ask them specifics about how they plan to support the customized software required for each and every ATCA solution they propose to sell to the market? Remember that no two solutions are alike.

That's why the big guns got out of the ATCA market - each ATCA solution required a team of software programmers, which is an untenable business model.
Hanover_Fist 12/5/2012 | 2:57:46 PM
re: HP Ditches ATCA (a) When you "own" the architecture, you control your own fate and investment - a concept that the networking vendors (i.e., Cisco IOS, Juniper JUNOS, etc) well understand. IBM, HP, Sun, etc... follow this same development path by building their own (read closed) version of a blade system.

(b) The jump from single threaded to multi-threaded is happening in a HUGE way, bolstered by both AMD and Intel. It's a different situation completely. The world is migrating towards multi-core/multi-threaded applications. Every ATCA solution must be custom built to accommodate a particular vendor's solution and targetted implementation. With Intel out of this market, where's the base of programmers with knowledge going to come from?

(c) There are NO widely accepted/adopted COTS API as you mention. This means that each and every solution using 'open' architectures must be hand built. What a waste of time and resources, especially when you multiply this effect for each and every ATCA component that could potentially come from a different vendor...just because it you can plug it into the hardware does not guarantee that the hardware will actually work as required (i.e., you MUST touch the software to make the hardware do what you want).

Why waste time/money/development efforts in this fashion? The Intel promise of lower component costs from the competitive nature of ATCA as COTS technology doesn't hold water because of the enormous costs of the software development cycle. This is the key point that the Embedded System Vendor market do not understand and don't get - they were/are dazzled by all of Intel's promises but with Intel out of the market (just like with the IXP), guess what's going to happen.

Programming IXPs was too complex and the time it took to get your head around the 'open programming architecture' turned out to be the same time requirement for building the software onto of your own hardware (read private ASIC or other reprogrammable component) - why waste the effort?

The whole ATCA argument posited to the HPs, IBMs, others avoided/ignored the software situation - it was presented as a 'if you build it (the ATCA hardware), they will come (the software programmers).'

Haha...what a joke. It's a nice idea on paper but in reality, unless you push on the hard questions (how many programmers do you ACTUALLY have building the internals), the products will never see the light of day at the promised 'low cost' points.
desicommguy 12/5/2012 | 2:57:46 PM
re: HP Ditches ATCA Hi,
I do agree that availability of pre-integrated COTS software is an issue but my questions are:
(a). Is this any different for whether the platform is ATCA or the blade architectures
(b). Isn't most of the software now have to be re-architected or re-ported to take advantage of the multi-core Intel/AMD blades for most of the control, management and embedded control purposes
(c). Also, there has been very little adoption of COTS software API's for control and management middleware let alone application solutions and isn't this more of an issue for any standard platform adoption

Especially, in the context of the HP comments their solutions will be a competing offering to the TEM's for applications and therefore not sure why the TEM's would adopt the HP Bladeserver architecture for their own applications?

Any thoughts and feedback would be appreciated.

Hanover_Fist 12/5/2012 | 2:57:46 PM
re: HP Ditches ATCA Yeah...and look what Intel did to all those chumps...err, I mean embedded system vendors who bought in to the grand promise of network processors (not to mention all those unfortunate programmers dumped during the latest rounds of Intel layoffs).
Pete Baldwin 12/5/2012 | 2:57:46 PM
re: HP Ditches ATCA Simon Stanley (our Heavy Reading analyst following ATCA) points out that IBM and HP will get economies of scale here -- since they're already producing BladeCenter and BladeSystem in volume.

ATCA can get there, but it will take years.

His conclusion: "I do not expect HP and IBM to dominate for carrier-grade applications in the longer term, although they may do so in the medium term."
opticalwatcher 12/5/2012 | 2:57:45 PM
re: HP Ditches ATCA rainbowwarrior said it best back in March:

paolo.franzoi 12/5/2012 | 2:57:45 PM
re: HP Ditches ATCA
2 Things:

1 - Using Intel and anything associated with embedded systems has been stupid since the day they MDed the 8085. NPs are doing fine, thank you and have a role in network equipment. If you have chosen Intel for an embedded system, you will hate yourself later (8085, Peripherals [8255,8257, etc]. 80186, 80386SX, i860, i437, i960, iXP, Level 1 stuff).

2 - If you are building custom hardware and software than ATCA or any other standard piece of hardware should save product development time. If you have a better "off the shelf" platform, go for it. But the only software savings, is simplicity of having working hardware at the outset.

Sign In