HP started then spiked HP-UX on x86 project
Juicy Itanium trial document firehose opened
As part of the ongoing lawsuit about whether or not Oracle had committed itself to supporting its software on Hewlett-Packard's Itanium-based servers, the software giant did a core dump of very interesting documents that show what many of us suspected: that HP did indeed mull acquiring the Sparc/Solaris business and that HP did in fact have a skunkworks that was porting the HP-UX variant of Unix to the x86 processor from Itanium.
It looks like at one point HP was also working on a big fat SMP server to run that HP-UX variant, and based on the Opteron processor from Advanced Micro Devices.
It is a bit surprising that Oracle made the documents – a collection of emails and presentations presumably obtained through the discovery process – public. The documents were presented along with a letter from Jeb Dasteel, senior vice president and chief customer officer at Oracle.
As is the case with any heated issue, there are many different sides to the Itanium story – at least three in this particular case, with HP and Intel's side being just two points of view and Oracle's being the third. You can also bet on it that Oracle was very careful about what documents it presented in its effort to show that HP and Intel were not exactly pleased about the way the whole Itanium thing had turned out and were arguing about who should pay what in the ongoing development and manufacturing of Itanium chips.
For those of us who enjoy the intense competition in the Unix server space, the two most important documents that Oracle outed relate to Project Blackbird (PDF), which was a rationalization for buying Sun Microsystems back in February 2009, and Project Redwood (PDF), a four-prong battle plan that HP cooked up in March 2010 – two months after Oracle completed the acquisition of Sun.
The Project Blackbird analysis reveals where HP's Business Critical Systems business was at this important juncture. Scenario A more or less conforms to the current reality except that it did not predict that Oracle would upset the Itanium applecart in March 2011 when it decided to cease development and support for its database, middleware, and application software on future Itanium hardware.
HP is for all intents and purposes the only Itanium server seller and its HP-UX Unix is entirely dependent on Itanium processors. In Scenario A, HP continues to develop Itanium systems with Intel and drives $30.8bn in revenues from Integrity and Superdome machines from fiscal years 2009 through 2014 (inclusive), with $1.4bn in research and development costs and a net profit of $12.8bn. This includes all hardware, software, and services revenues in any way tied to an Itanium system.
In Scenario B, Itanium development stops with the quad-core "Tukwila" Itanium 9300 processors and revenue over those six years drops to $22.8bn, while HP does $500m in R&D, and brings $9.2bn to the bottom line. If you do the math, the margins are a bit higher in Scenario A, but not by much.
HP-UX 'on a death march'
The interesting thing is that, either way, this BCS business is in decline, just like the Unix business at Oracle is in decline. It is worth noting that even IBM knows it cannot sustain growth in the Unix space much longer and is trying hard to expand the Power Systems franchise into areas where X86-Linux play well.
HP liked the synergies between HP-UX on Itanium and Solaris on x86, and added that Solaris was the leading Unix in the United States with HP-UX being the leader elsewhere. There could have even been some play with the Zettabyte File System running on ProLiant servers to make "industry standard" storage arrays.
The most damning page in this presentation – it is not clear who gave it – lays it out plainly enough, with page 6 saying that "HP-UX is on a death march due to inevitable Itanium trajectory" and adding that 45 per cent of the revenue and 60 per cent of the gross margins in HP's Technology Services group come from HP-UX systems and "we do not have a go-forward."
The Project Blackbird presentation goes on to say that x86 is "on a credible trajectory to fulfill all aspects of RISC within 5 years," that HP is not going to own its own "software IP stack" on which to "build value" as hardware gets commoditized, unlike rivals Cisco Systems, Microsoft, VMware, and NetApp. (HP forgot to mention IBM, which has lots of software stacks, not just one.) Buying the Sparc/Solaris business from Sun (or perhaps all of it, it is not clear what bits HP was looking at) would give HP a Unix on x86 iron with strong ISV support and would make it "a two horse race in the enterprise: Solaris and Windows (which both run best on our platforms)" and HP would move Sparc blades into the BladeSystem c7000 chassis much as it has done with Integrity and Superdome 2 machines, and has promised to do with this year's Project Odyssey.
HP was also contemplating the fact that it could use Solaris in switching products and added that Cisco has dependencies on Solaris (this may not be news to you, but it is to me), and worried about what would happen if IBM got its hands on Sun. HP was worried that Big Blue would consolidate all of its efforts onto a single operating system (presumably Solaris, but maybe AIX) and use the Transitive QuickTransit emulator that it bought back in November 2008 to emulate all other apps running on all other platforms. HP said such a deal would saves its x86 server business "from disaster," give IBM storage cred and control of Java, and isolate HP-UX.
That last bit pretty much describes what Oracle has done by buying Sun and then dropping the iceberg in front of Itanic.
Blackbird singing in the dead of night
On March 3, 2010, Martin Fink, general manager of HP's BCS unit, gave a presentation to Ann Livermore, who was running Enterprise Services at the time, and Dave Donatelli, who still runs the company's Enterprise Systems, Storage, and Networking group in which the BCS unit sits.
The interesting bit in that Project Redwood document from Donatelli to Livermore, dated January 14, 2010 – a week before Oracle closed its acquisition of Sun for $7.4bn (net of cash, around $45.6bn). Take a read:
I am waiting for the price tag and schedule on the x86 roadmap. lt will take about 30 days for the team to complete the detailed work. Our first milestone was hit when the team could boot HP-UX on x86 on 12/23. I am expecting an incremental 100 million plus price tag to do the port. As i mentioned before, the engineering road map when I arrived called for the exit from the business. Based on work that Vincent and I have done we have determined that the combined BCSITS earns more than our PC Business. Yesterday I asked NEC to help fund the port. They were quite pleased to learn that we planned on staying in the business for the long term."
HP's Project Redwood options
The plan that Fink put together for this presentation is incredibly detailed, as it was the one the top managers would be using to make a decision to invest in commercializing a port of HP-UX to x86-based servers and the impact that it would have on the server racket.
The plan required $489m in net investment in fiscal years 2010 through 2014, with x86 Superdome blades (which are under development today through Project Odyssey) costing $255m; the HP-UX port costing $147m; other software such as hypervisors and clustering code costing $24m (some of this is also being done under Project Odyssey today); and performance and ISV testing costing another $128m. HP was also budgeting $70m a year for Intel – to pay for it to fab Itanium chips in fiscal 2010 through 2013.
HP was being fairly aggressive in its plan to get HP ported to Xeon chips and ramp up in the 1+2 plan, as Fink called it. Here's what the server roadmap looked like:
And here's what the HP-UX on x86 server release schedule looked like:
As you can see, those Superdome x86 servers that HP talked a bit about last November were expected sometime around June 2010, in this scenario. HP kept Itanium alive through the "Poulson" and "Kittson" generations out to 2016 with support for a decade after that based on HP's past history.
The HP-UX rollout plan under Project Redwood had updates to HP-UX 11i v3 for Itanium machines in both 2012 and 2013. The developer release for HP-UX on x86 was scheduled for early 2011, and release 1.0 was set to offer basic hardware support as well as the ability to run it within a VMware ESXi hypervisor virtual machine partition. Later in 2012, the HP-UX for x86 machines would be available on the Xeon-based Superdome machines and in early 2013 it got real interesting with an enterprise release and an application binary interface (ABI) runtime that was compatible with Red Hat Enterprise Linux.
This sounded like a pretty good plan. But one of the several CEOs or GMs who oversaw HP between 2010 and now took a chainsaw to Project Redwood. And if that person has not been fired, maybe they should be part of the layoffs that are expected to come this week at HP.
Here's why. If HP did not somehow preserve the HP-UX revenue stream and just rode out Itanium to the future Kittson processor and did nothing to port it to x86 iron or license Solaris or do something else, in order for Donatelli's ESSN group to hold profits flat from fiscal 2012 through 2017, HP was making a certain projection. It was projecting that its ProLiant and BladeSystem business would have to grow at twice the market rate, that its networking business would have to grow at twice the market rate, and its software business would have to grow at nearly four times the market rate. And just riding Itanium down would be price pressure on HP and benefit margins at Oracle and IBM. This is by HP's own admission.
The third plan, which was also a part of Project Redwood, was to take an easier way out and license Solaris and push it on ProLiant iron. This was just fraught with issues, some of them emotional and hence unappealing – and certainly not possible once Oracle seceded to stop development on future Itaniums.
The fourth option in the Project Redwood analysis was to think about baseball and prolong the Itanium lifecycle as long as possible, literally paying Intel $218m to make the Kittson Itanium and another $218m to do a kicker to this chip in 2016. This plan also included getting socket-level compatibility for Itanium and Xeon processors, which is expected as part of the Project Odyssey Xeon-based Superdomes but which has not been confirmed by either HP or Intel.
HP is in a very tough place and has been since it was obvious that Itanium was not going to take over the server world, let alone the desktop. Even if HP did manage to get HP-UX on x86 running (not just booting), once Oracle bought Sun, the odds that Oracle would port its code to this HP-UX on x86 were pretty small.
Oracle already owned a widely regarded Unix that ran on x86 iron – Solaris – and had its own Linux for those who prefer that and Sparc/Solaris and Windows as the final fallbacks. HP-UX would be just another operating system outside of its own control that cost it money to support without necessarily bringing Oracle incremental revenue. Oracle knew this, and HP knew this, and this was probably why HP put HP-UX on x86 in the wood chipper and went with the easiest way out.
That would be Project Odyssey, which means putting Xeons into Superdome iron and beefing up Linux and Windows to be as reliable as they can be while peddling Itanium-based machines as long as possible. And that is just not as much fun as it could have been. ®