Turning a corner with the new Itanium

The Itanium-based Integrity server line is fully fledged, the software is in place, and the new "Montecito" model doubles performance over the last Itanium. Now HP's Itanium server revenue is growing, as customers move away from machines using HP's older PA-RISC and Alpha chips, said Rich Marcello, general manager of the company's Business Critical Server group.

"We're at 38 percent of revenue through this last quarter," he said. "We'll be at 70 percent next year, in fiscal 2007. That's our plan."

That transition has been slower than expected at HP, which conceived of the chip family but relied on Intel to bring it to design and manufacture. Delays, poor initial performance and software incompatibilities hampered Itanium's debut. In addition, Intel delayed the arrival of Montecito from 2005 to the second half of 2006. Montecito, the first dual-core Itanium, promises to roughly double server performance.

"I think things are progressing, but HP has been consistently more optimistic than has turned out to be justified," said Illuminata analyst Gordon Haff.

Marcello makes no apologies for what he's done over the three years he's led HP's Itanium work. He prefers to direct attention toward Itanium's advantages over competing RISC (reduced instruction set computing) chips such as IBM's Power and Sun Microsystems's Sparc. For example, several server companies use it, Intel's manufacturing clout backs it up, and it can run multiple operating systems, including Windows, Linux, and HP's OpenVMS and HP-UX.

Marcello sat down with CNET News.com's Stephen Shankland to talk about Itanium servers and the road ahead.

Q: You have new Integrity products coming out with Intel's Montecito Itanium chip.
Marcello: We are completely refreshing the line, all the way up to our Superdome servers. We're also adding two new servers, the 3600 and 6600, which are midrange servers. We've been waiting to get Montecito now for a while, so there's a lot of pent-up demand.

How aggressive are you with pricing?
Marcello: We're not going to be raising the prices. The pricing will be very aggressive versus our competitors, IBM and Sun. When our customers see not only what we're offering in terms of initial system sales but also in terms of the three-year total cost of ownership, I think they'll be pretty pleased.

What lowers customers' effective cost over three years, compared with the initial acquisition cost?
Marcello: Our Montecito servers get twice the power efficiency. Whether they're implementing blades or Integrity servers or ProLiant, we optimize the power across the board. And with virtualization software, we believe that if they're consolidating down from many, many servers, they're going to actually get some cost-of-ownership benefits there as well.

HP IT (department) is going from mega datacenters down to six. We're buying thousands and thousands of Integrity servers.

One (way HP is using them) is a shared application server, where they're going to stack applications on top of our HP Integrity servers. If you're a guy who owns a server and an application, you want to make sure you maintain control over that. It's kind of like when you're a kid: You don't want to share with your brother or your sister. But (when using partitioning and virtualization technology) you can guarantee with Workload Manager or Global Workload Manager that you're going to give that user the same level of resources. You get all the benefits of consolidation, but you also get the benefits of the user feeling like it's his or hers.

This is the phenomenon of server hugging, where nobody wants to lose control over his or her department's hardware. But it still requires centralized IT purchasing and managing severs, so do departments still worry about losing control?
Marcello: That's right. But the benefit is that you save a bunch of money, and for the most part--at the 95 or 98 percent level, you're getting the same level of server control as you would have gotten if you owned the server yourself.

When HP reported earnings for the most recent quarter, Business Critical Systems's revenue decreased 6 percent overall. Although Itanium Integrity grew 76 percent, it was more than offset by Alpha and PA-RISC server declines. Would that have happened if Montecito had been on time?
Marcello: I don't know. It's a hypothetical. I do think that there's a lot of pent-up demand for Montecito, and so my overall view is that it is highly likely that customers are waiting for it, and so that could be true.

What was the effect of that Montecito delay on you?
Marcello: I think we would have preferred to have it earlier. I think that we mitigated a lot of it by introducing our (high-end) Arches-based systems back in the springtime.

And after 2011?
We renegotiate. We extend the agreement. There's nothing magical about 2011. You do these things five years at a time, basically.

You mentioned your competitors, Sun and IBM. For a long time, HP and IBM have been saying, "It's a two-horse race," not even mentioning Sun. In the second quarter, Sun was the only one of the big four server companies to show revenue growth. Is this now back to a three-horse race?
Marcello: I don't think so. First of all, that was Sun's fourth quarter. If you look at their Q4, our Q4 or IBM's Q4, it always tends to be the best quarter, so you probably could make that claim about any of the three of us in that quarter. Second, if you look at what's happened over a multiyear period, at some point, you'll have a decent (year-over-year) comparison.

Sun's giving a lot of attention now to open-source Solaris. What do you think of that strategy, and is it helping them at all?
Marcello: I think the open-source strategy is not something that we're going to go forward with, from an HP-UX perspective. I think it's also something that our competitors at IBM said they're not going to go forward with as well. You have to look at what folks are doing in terms of downloads, and what they're then doing in terms of buying additional software and support from Sun, versus what they would just buy out of the chute, then look at the total cost of ownership of the two. I think when you look at that analysis, it ends up not really being that much different.

Does it make any difference in popularizing the product?
Marcello: Only if people actually use it more. In terms of us going forward with a strategy where we would want to do something similar, we just don't think it makes a lot of sense right now, that's all.

Back in the debut years, HP and Intel argued that Itanium is the architecture that will replace RISC. Itanium has the legs to go for two decades. Do you still believe that? RISC in general still seems competitive.
Marcello: I don't think that IBM will stop investing in Power any time soon. There will be one other player for sure.

I think the target markets (for Itanium) are RISC replacement, and as a mainframe alternative and high-performance computing. If you take the server market and split it into two, Itanium is really focused on the upper $25 billion.

What fraction of customers mix and match different operating systems on one Integrity system?
Marcello: It depends. Right now, not many. As we move more and more toward...consolidation, we expect more and more customers to run multiple operating systems.

With Microsoft changing strategy to focus only on high-end workloads, like back-end databases for Itanium servers, it doesn't strike me as terribly likely that people are going to take one Itanium server and split it up into 15 low-end Window servers. So, is Itanium going to be good for consolidating lots of little underutilized Windows servers?
Marcello: It would be good for consolidating lots of Windows SQL Server. The real focus here is around database and business intelligence and SAP kinds of things. I think there you can definitely consolidate.

I'm not saying you can't consolidate databases. I'm just saying that the majority of consolidation arguments I hear involve low-end machines like file and print servers, that's not really the market for Microsoft and Itanium.
Marcello: Yeah, and I think the market here is about kind of SQL consolidation really.

For high-end Itanium severs, you've got the Arches chipset that's going to last through Montvale (Montecito's successor, due in 2007). What comes after that, with Tukwila systems?
Marcello: We've got Windjammer (successor to the Arches chipset) coming down the road.

How about on the low end?
Marcello: Yeah, there'll be a follow-on as well. It's something called Corsair.

What kinds of changes will come with that generation?
Marcello: It's a brand-new design--everything that you could possibly want, in terms of increased availability, scalability, robustness around partitioning, the true ability to build clustered failover systems. Anything that you can think of that would add to the value proposition around flexible capacity, simplified management, security and availability, we're going to throw in there. It's going to be a big deal.

Do you see multiprocessor systems with more than 64 sockets? (Superdomes today can accommodate as many as 64 processors.)
Marcello: It depends upon what customers want. Certainly 64 sockets is reasonable. We have the capability to go higher if we choose to do so, but my hunch is that we'll probably stay in that general range.

Is Poulson, the Itanium model three generations down the road, going to be a major jump after Tukwila, which is two generations away?
Marcello: Yeah.

Will that be another doubling of performance?
Marcello: That would be good.

That's what you're hoping for, or that's what you're betting on?
Marcello: That's what I'm hoping for. You know, I think it's still four or five years out.