HP Itanium's Ignominious Demise (Updated)

I thought I'd let the dust settle a bit before commenting on recent news concerning the fate of HP's Itanium servers. I've worked with some organizations that have purchased a lot of Itanium servers, and I know they're all very distressed with software vendor abandonment of HP's flagship platforms.

First, a brief review of history. In the 1990s, HP still designed its own CPUs for its high-end servers, notably the PA-RISC architecture for its UNIX servers running HP-UX. However, HP's homegrown CPUs were rapidly becoming uncompetitive as other vendors (including IBM) could sustain bigger CPU R&D investments. HP had also acquired responsibility for former DEC platforms (specifically servers running the VMS operating system) and the Tandem NonStop platform (which some analysts classify as a mainframe environment), among others. In an effort to cut costs, HP decided to end support for some operating systems completely, deliver minimal new function for others ("maintenance mode" only), and consolidate its operating systems onto a single CPU, with most of the design and fabrication outsourced to share costs. Thus HP struck a partnership with Intel to jointly create the Itanium CPU. The forecasts were rosy, and Itanium would allegedly take over the world. In particular, Intel agreed that its X86 architecture would remain forever 32-bit only, while Itanium would be anointed as its sole 64-bit architecture, with a new instruction set. For migration purposes, Intel and HP agreed to equip the Itanium CPU with a 32-bit X86 compatibility mode. And (HP promised), while its customers could experience some disruption moving their applications from other CPUs to the new CPUs, that would be a one-time problem and (HP claimed) worth the effort and expense.

That was the plan. But almost nothing went according to plan. Itanium was always late and slow. AMD created and shipped the X86-64 instruction set (called AMD64) in its CPUs, forcing Intel's hand in doing exactly the same thing. All but a few tiny vendors stopped shipping Itanium CPUs, leaving HP as the only company shipping Itanium CPUs in large numbers. And IBM kept gaining high-end server marketshare from HP, becoming far and away the #1 UNIX server vendor. IBM also nurtured the mainframe's renaissance.

Even so, for many years HP managed to keep a respectable Itanium business going, with HP-UX emerging as the #2 UNIX platform as Sun collapsed in the 2000s. But then some other shoes dropped. Microsoft stopped developing software for Itanium, eliminating Windows Server and SQL Server from the platform. Red Hat stopped shipping new versions of its Linux distribution, which also meant that Red Hat's JBoss application server left the platform. Those software vendor announcements were bad but probably not fatal. But then HP's Board of Directors fired the company's CEO, Mark Hurd. Hurd, a personal friend of Oracle CEO Larry Ellison, quickly joined Oracle. Oracle, fresh from its acquisition of Sun (and the #3 UNIX server vendor), which as recently as 2006 had been HP's (and HP-UX's) strongest and most profitable software vendor, announced in late March that there would be no new versions of any Oracle software products for Itanium. Oracle Database 11gR2 would be the last — although HP customers are free to pay Oracle a perpetual stream of support fees for 11gR2 if they wish.

Needless to say, HP (and to a lesser extent, Intel) were stunned. But nobody buys servers only to fill racks in data centers. Companies and governments buy servers to run applications and databases. HP's biggest and most popular Itanium-compatible software product — its only big "anchor tenant" — has now hung up the "going out of business" sign and is closing its doors. Ironically, IBM is now Itanium's largest software vendor, by far. Also last month (in case the message wasn't clear enough), Oracle shipped the latest Oracle Database 11gR2 for Linux on IBM System z.

Let me state my assessment plainly: Oracle's announcement is Itanium's death sentence. It's also brutal but highly effective business strategy for Oracle in its competition with HP. Oracle will probably lose some HP customers who are (rightly, in my opinion) concerned about doing business with a vendor that abandoned them, but Oracle is betting, correctly, that most of its customers, when forced to choose sides, will choose Oracle's software over HP's hardware. To paraphrase a former U.S. presidential advisor, "It's the software, stupid."

For Intel it's a short-term annoyance but nothing more. This week Intel proceeded with its latest Xeon announcement and paid mere lip service to Itanium, at best. Intel would be quite happy to save R&D expenses and focus exclusively on X86 and, in particular, try to expand into mobile device markets and do battle with ARM. However, Intel would prefer not to anger its biggest OEM (HP) and might face some financial penalties acting on its own, so Oracle's announcement is in some ways a nice alternative.

And then there's IBM, which is extremely well positioned with its Power, System z, and software offerings, to attract and win the many soon-to-be-former HP customers. IBM can assist HP Itanium customers in any manner they wish: in straight-up migration from the #2 UNIX platform to the #1 (and growing) UNIX platform, in enterprise consolidation (through Power and/or System z) for additional cost savings and operational benefits, and/or in tactical or strategic migrations from Oracle software products to IBM software products, such as Oracle Database to DB2 (even on HP-UX), immediately or over time. Oracle's announcement also helps validate IBM's entire business strategy over the past decade-plus (and invalidate HP's): "It's the software, stupid."

As for HP, I think their new CEO has many challenges. There are some strong technical reasons (big endian v. little endian) why HP will find it very difficult to move HP-UX and other Itanium operating systems to Intel/AMD X86-64, and even once that's done (if it can be done) Oracle's software is still not going to support HP-UX. I suspect HP will just have to muddle through the Itanium meltdown and, as Intel improves X86-64, move toward an Itanium emulation solution for any remaining VMS, NonStop, and HP-UX customers. That'll provide application compatibility but not performance. Alternatively, HP might want to call IBM to inquire about a move to Power and/or System z CPUs which would be technically more suitable for HP's operating systems and which could at least be partitioned and virtualized to run other operating systems with continuing access to new business software from many vendors. Otherwise, I think HP is fated to devolve into much more of a Dell-like commodity X86-64 server (and PC) vendor, of course with lower profit margins. I don't know how HP builds a software business essentially from scratch. HP's business software portfolio is very limited, and there aren't too many affordable software vendors available for acquisition. Some analysts have suggested that HP should buy EnterpriseDB and try to migrate Oracle Database customers there, but I don't think that strategy could possibly work. In short: well played, Larry.

UPDATE #1: Expanding on the "HP on IBM" idea for a moment, IBM acquired Transitive, the company that produced Apple's superb Rosetta technology which helped Apple (ironically) move from PowerPC to Intel CPUs while maintaining compatibility with binary PowerPC applications. That technology works very well, and IBM Transitive would be the best in the world in helping HP's customers transition from Itanium to Power and/or System z CPUs while maintaining binary compatibility. Also, technically IBM and HP could do something very similar to what Apple did and place that CPU emulation within the same operating system instance. That would be highly desirable for HP-UX Itanium binaries specifically, which could then run as-is within AIX. (VMS and NonStop are probably too "alien" and would need to run in separate LPARs or virtual machines, but those two operating systems don't run inside HP-UX today, so that's fine.) Of course, HP would have to contact IBM, strike a deal, and collaborate. Given that IBM is now HP Itanium's largest software vendor, and that IBM possesses the best set of technologies to help HP customers migrate (whether HP collaborates or not), I think HP's CEO probably ought to give his counterpart at IBM a call if he hasn't already. It's better for HP to save some revenue, profit, and customer relationships. There's ample precedent for that sort of partnership between (former?) server rivals. For example, Hitachi is a major IBM Power and AIX OEM, and Hitachi and IBM also collaborated on the z800 mainframe.

UPDATE #2: Oracle's announcement also severely and negatively impacts HP's remaining VMS and NonStop customers. Oracle owns Rdb, the DEC-created database engine that's probably the most popular database for VMS systems. Oracle also owns Tuxedo, probably the most popular transaction manager for NonStop systems. Oracle is halting development for these middleware products, too. VMS and NonStop customers typically have very demanding quality of service (QoS) requirements, including application stability and longevity, so my recommendation would be (in most cases) that they plan for and execute a migration to IBM System z. That won't necessarily be easy, but it's the best available option given Oracle's actions.

by Timothy Sipples April 6, 2011 in History, Systems Technology


TrackBack URL for this entry:

Listed below are links to weblogs that reference HP Itanium's Ignominious Demise (Updated):


Creating value for society is an integral part of our business. We believe that contributing to the well-being and development of society is an extension of everything we do. Reaffirming our role as a contributing member of social and economic environment,
we align our business operations with social values.As a responsible corporate citizen, we leverage the power of IT to provide the platform for young rural enginers to succeed and prosper in their life,providing free training on mainframe Technology for more info you can visit www.gowdanar.com

Posted by: satya | Jul 15, 2011 7:54:52 AM

You have no knowledge of Historic events regarding the development of the Intel/HP Itanium processor, nor do you have knowledge of HP acquisition of Tandem, you are biased towards IBM, thus your paper has no credibility. Check you facts as your readers maybe informed professionals.

Posted by: Hector Gull | Feb 6, 2012 1:02:17 AM

I have worked in operations with both Oracle on X86 and NonStop SQL on Itanium, NonStop is the most secure and problem free in my 24 years hands on in the field, it may cost a fraction more, but is up and running, with no history of hackers attacks or security issues 99.99999% with mostly downtime for maintenance reason, no issues with the Itanium that I can recall as a "BIG FAILURE", they actually hardly fail.

Posted by: Hector Gull | Feb 6, 2012 1:12:29 AM

The comments to this entry are closed.

The postings on this site are our own and don’t necessarily represent the positions, strategies or opinions of our employers.
© Copyright 2005 the respective authors of the Mainframe Weblog.