HP Preps for PA-RISC to Itanium Transition

HP will sell PA-RISC through 2006 and continue support through 2011.

• By Stephen Swoyer
• 10/08/2002

At the same time, says John Miller, director of server marketing for business critical systems with HP, the company will nevertheless release two additional iterations of its PA-RISC chips. “If you look at our roadmap, we have two more revs [of PA-RISC]—PA-8800 and PA-8900—which will take us out approximately through 2006.”

The goal, Miller says, is to make the prospects of a PA-RISC-to-Itanium migration as painless as possible for HP’s existing HP-UX customers. For example, he points out that customers who purchase new PA-RISC 8900 systems in 2006 will enjoy end-of-lifecycle support through 2011.

“There’s a lot of time for customers to [transition]. We’ll support them for five years after they come off our price list, so assuming they come off our price list in the 2006 time frame, we’ll support them for another five years after that, through 2011 roughly,” he acknowledges.

To assist customers who may be undecided about moving to Itanium, HP next year will introduce a new chip—the PA-RISC 8800—that exploits server hardware that is capable of supporting either PA-RISC or Intel’s next-generation Itanium chip, code-named “Madison.”

“These will enable board upgrades to our existing PA-RISC product. So you literally swap out your PA processors and processor boards, and you swap in the new [IA-64] processor boards. This gives our customers the flexibility to move to [IA-64] when they want to.”

Unlike the ClearPath Plus and ES7000 systems marketed by rival Unisys Corp., however, customers will not be able to configure HP’s servers with mixed proprietary and Itanium processor architectures. Unisys, for example, simultaneously supports its own proprietary CMOS along with Intel’s IA-32 and IA-64 chips in its ClearPath Plus mainframes. Instead, Miller concedes, HP will offer customers the ability to run either PA-RISC or Itanium—but not both at the same time—in its high-end servers.

“Technically, there’s no reason why we can’t partition, because the [IA-64] board will also support our PA-8900 processor, [which is] our next generation processor,” he observes. In practice, Miller continues, “HP is choosing not to make that investment” primarily as a result of customer feedback.

HP introduced a 16-way Itanium server, the RX-110, in 2001, but has thus far concentrated largely on marketing two- and four-way Itanium 2-based systems. In contrast, rivals Unisys and NEC Corp. have demonstrated 32-way Itanium 2-based systems. For his part, Miller claims that HP-UX customers aren’t demanding high-end Itanium 2-based systems. Instead, he says, customers “will just want to buy a four-way Itanium system, test it out, and make sure that it works with [their] applications” before they commit to the IA-64 architecture on such a large scale.

Nathan Brookwood, a principal with microprocessor consultancy Insight64, says that HP’s strategy is in large part dictated by the limitations of the current Itanium 2 “McKinley” microprocessor, which he says features an insufficient L3 cache for scalability in large SMP configurations.

“Their strategy, I think, is dictated by the fact that the current Itanium [2] is only available with 3 MB of L3 cache, and that greatly precludes the opportunity for performance in large systems in high-end SMP configurations,” he points out. “Next year, when Madison comes along with larger L3 caches and can begin to scale, HP will move to the larger configurations, so it makes a lot of sense.”

Starting next year, HP’s Miller confirms, his company will introduce Itanium 2 in its high-end servers, including its 64-way SuperDome systems.

“We are planning on rolling out a Madison-based system in the higher-end parts of our product line, [such as] our eight-way system, our 16-way system and on up to our [64-way] SuperDome servers,” he confirms.

Rob Enderle, a senior fellow with consultancy Giga Information Group, says that HP’s Itanium-to-PA-RISC transition strategy goes a long way toward addressing some of the potential concerns that customers may have about the future of their investments in HP-UX. In this regard, Enderle suggests, HP is particularly concerned about losing existing HP-UX customers to its Unix competitors, particularly Sun Microsystems Inc.

“[HP] committed early on to delivering a version of HP-UX that was tuned for Itanium and [which] was a feature-for-feature match [with HP-UX on PA-RISC],” he points out. “So they’re making a monumental effort to make it as simple as possible to move between the two architectures. Their hope is that they don’t lose a lot of people in the migration, [because] every time you do a migration like this, a significant percentage of your customer base will take the opportunity to explore other vendors.”

Alpha Futures

HP has also committed to delivering two additional iterations of the Alpha microprocessor it picked up when it acquired Compaq Computer Corp. Considered by many to be among the finest microprocessors ever developed, Alpha anchored Compaq’s line of OpenVMS minicomputers and Tru64 Unix servers. Compaq itself acquired Alpha when it purchased Digital Equipment Corp. in early 1998.

HP is currently reselling systems based on the Alpha EV69 chips, but will introduce Alpha EV7 and EV79 chips over the next few years (through 2006). The company hopes to transition existing Alpha shops to Itanium as well, but it hasn’t yet introduced a hybrid hardware architecture similar to that developed for its HP-UX servers.

“There are several iterations of Alpha that will hit the streets, so it’s definitely a very orderly transition,” comments Insight64’s Brookwood. “By sometime in the 2005 or 2006 time frame, if HP’s transition plans work, most customers are going to be migrating over [to Itanium], and it will be a very natural kind of transition where they just won’t be interested in the proprietary platforms anymore.”