ITANIUM: Intel’s 64-bet

Does the 21st century signify a new era for business computing? Intel says it

does. It launched the Itanium processor in mid-2001, along with a roadmap for

its IPF. That’s the Itanium processor family, with the IA64 architecture.

The Itanium is Intel’s first 64-bit processor, to power medium and high-end

servers. It will support current Unix flavors, Windows 2000 and Linux, and more

powerful versions optimized for the IA64’s larger word size and a powerful

mode of parallel computing. Future members of the IPF will be 64-bit chips like

McKinley, Madison and Deerfield (all project code-names), carrying over into

2004.

The Itanium was not really Intel’s baby. Project Merced’s predecessor was

born on Hewlett Packard’s drawing-board in 1994. And the apparent ownership

change marks a visionary move in infotech history.

Through

the 1990s, HP shipped its mid- and high-range servers with its 64-bit PA-RISC

processor, the PA-8000 series. But continuing to push up chip performance would

draw its resources into areas difficult to sustain–beyond core competence.

Yet the processor is a big differen—tiator among Unix servers. IBM, Sun and

Compaq were shipping servers with their own, robust 64-bit processors like the

PowerPC, UltraSPARC and the Alpha. Processor innovation and performance gain was

a key to survival. HP decided it wasn’t going to do this on its own.

Intel Corp’s VP of sales and marketing, Daniel Russell, credits that to HP’s

long-term vision. "They realized early on that a top-to-bottom approach is

not sustainable," he says. "That’s where you do everything, the

processor, chipset, OS, drivers, apps, services." And it was always HP

strategy to develop a follow-on to RISC, according to Peter Hall, HP’s APAC

marketing manager for Unix systems. "We knew that someday, RISC would start

flattening," he says.

Eight years ago, HP found Intel a good partner for the long haul to build a

successor to the PA-RISC family. The Santa Clara, California-based chipmaker had

the economies of scale in fabrication and research, with the higher volumes and

lower costs that HP wanted. HP had the technology learning of building 64-bit

processors in the high-end Unix server segment–where Intel had not ventured.

The need to build a successor to PA-RISC–and perhaps to the Power PC and Alpha

as well–was compelling for both. The rest is now history.

Open vs Sun?



The same reasons could be the purveyor of doom for proprietary processors.

"For years we’ve been waiting to see if a proprietary system can go

open," says Intel’s Russell. "But we are yet to see one."

"HP was the first off the block with Itanium systems, but it will continue to support PA-RISC right through 2010 and beyond" Roy Vandoom, chief marketing officer for business systems, HP

What blocks a product’s shift from proprietary to open? For an established

proprietary environment, it’s the market relationships. Russell cites the case

of Apple which has never been able to open up the pieces that matter. The ISV

and developer community, retailers and distributors and user groups resist

changing the proprietary eco-system to an open-source and dynamic one.

"Only open standards will survive in this industry", says Narendra

Bhandari, Intel’s APAC Regional Manager (strategic relations)–Internet

solutions group.

As a result, vendors are often unable to innovate and transform at speeds

that match the needs of a competitive environment. With a proprietary product,

user volumes are smaller than a product with a more open community of developers

and users. The cost of innovation, spread across a smaller market, is higher.

The vendor often supports most of the cost, and the pace of innovation is often

low, leading to weak customer excitement and growth levels.

Intel’s product development strategy, on the other hand, has always

included a horizontal eco-system. Market relationships work toward increasing

the size and acceptance of the community, broad-basing and reducing the cost of

innovation and development.

Sun Microsystems, which has a relatively more closed eco-system, has to work

at not only the CPU, but middleware, OS and associated areas. "Sun has just

too many pieces, and they can’t keep up", Russell says. "Sun just

doesn’t get it", remarked Roy Vandoom, HP’s chief marketing officer, at

an Itanium summit for developers at Bangalore. He was referring to Sun’s

stonewalling on porting their Solaris OS to the Itanium platform. HP’s Peter

Hall is even more vocal, and says that product strategy of Sun and IBM is to

"lock in customers and lead them to a brick wall".

IA64’s slow build up



The IPF is Intel’s entry into systems for high-end apps, in the realm of

high availability, reliability and scalability. Intel expects Itanium servers to

be cheaper than RISC servers for the same levels of work management.

Improved work management means a server where apps rarely crash, one that

scales in terms of adding on CPUs during overloads, and which can compute

faster, for increasing database sizes. The Itanium processor has been designed

for all this from the ground up, Russell says, unlike Intel’s 32-bit

processors that were never intended to compete with RISC.

As you’d expect, HP was the first OEM to announce Itanium server and

workstation products, and a new OS. The new 64-bit HP-UX 11i OS has been rebuilt

from its legacy RISC OS, and is backward compatible with older HP-UX apps. This

first off-the-block achievement, capping off its head-start from 1994, has left

the HP camp with a sense of relief. But HP doesn’t under-estimate what lies

ahead. Says Vandoom: "We’re off to a great start, but it’s a 15-year

journey." The Itanium family was visualized as the successor to HP’s

PA-RISC family. With the last PA-RISC processor launch slotted for 2004,

combined with another five years for the product life cycle, HP will support the

PA-RISC platform till beyond 2010. That should be long enough for even the most

loyal RISC customers to migrate to the Itanium, says Vandoom.

IPF system shipments are expected to dominate the pie by the end of this decade, as RISC offerings fade away (HP-Compaq), or are further squeezed (Sun)

HP has obviously benefited from its early start and its development of EPIC

(explicitly parallel instructional computing) for the IPF. But Intel has also

done its best to ensure a level field for other players in its horizontal

eco-system.

But nobody is talking of volume shipments of Itanium systems, yet. No big

numbers, common with new 32-bit chips. Well, there’s is a hitch in the Itanium

rollout, and in the power it can wield in the initial stages. Neither the

current Unix and Windows flavors, nor the business apps, are good for it without

modifications. So neither Intel nor HP is looking closely at the numbers right

now.

And of course, there’s a basic issue. The Itanium’s raison d’etre is

not to convert 32-bit Intel systems users. It is to convert high-end RISC-UNIX

systems users to ‘Intel Inside’. Given the association of ‘Intel server’

with ‘PC server’, that’s an uphill task.

Hazy transition



The intermediate transition of business apps from Intel 32-bit and RISC

64-bit server platforms to IPF is not easy, nor the outcome predictable. For a

smooth transition, any OEM needs to provide for new OS supporting apps from

previous versions, packaged apps from ISVs, compilers for home-grown apps, and a

migration path for legacy apps. With Windows and Linux 64-bit OSs still in the

beta stage or just beyond, and only a few business apps ready, commercial

shipments of Itanium are still several quarters ahead.

Itanium wasn’t intended for commercial shipments. Meant to indicate the

start of the IPF life cycle, this first IPF family-member has not been marketed

as a volume product. "It’s aimed at the early-adopter minority and the

developer community", says HP’s Peter Hall. With the large base of

software developers in the country, India is therefore an important market for

Itanium.

Says Gartner analyst Matthew Boon, "The message is that the Itanium is

preparing for what is going to happen." The real test of the platform will

start when McKinley, next in the IPF line-up, is released later this year. Thus,

Intel is not surprised that shipment numbers have not been soaring from the word

go.

The facilitators for IPF adoption therefore depend on the availability of OS

and business apps, and migration paths. Forecasts for server shipments from IDC

indicate significant shares for Itanium shipments only from 2003 or 2004

onwards. In 2004, one out of every four servers shipped will be IPF-based.

Moreover, shipments of RISC and Intel 32-bit server platforms will continue till

2010 and beyond with IPF servers gradually increasing their share.

ISVs cast the dice



Several server architectures will be available during the Itanium’s

ramp-up. IA64 acceptance depends on several factors. The big one is the porting

of apps by ISVs from other platforms to Itanium. HP’s Peter Hall describes

this as a five to seven year migration period in the Itanium family’s total

lifecycle of over 20 years. There are over 150 OEMs and ISVs developing system

platforms, workstation solutions, enterprise apps, compilers, tools and OSs for

the IPF. Intel’s Bhandari says that roping in the next 500 partners is going

to be the real challenge influencing the rate of adoption. While OEMs (original

equipment manufactures) like IBM and Compaq may finally complete an improved OS

for their IPF products, just the OS will not be enough. As HP’s Vandoom says,

"Windows NT was ported on the Alpha chip for Compaq as well as on the

PowerPC chip for IBM’s AS/400, but was that successful?" No, thanks to

the absence of business apps there. Says Kit Maloney, IBM’s business manager

for X-series servers, "All leading vendors will offer Itanium boxes, but

not many customers are likely to have applications. And that’s okay for any

new generation chip". Still, user migration will take place when apps are

present.

“IBM was the first to deploy the Itanium on both server and workstation products. These are development platforms, as not many apps are available yet. In fact, we have already announced an IBM product roadmap on the future McKinley chip” Abraham Thomas, MD, IBM India

Redmond vs Linux



So which ISV has the sheer power to build the momentum for user migration to

Itanium? The answer is obvious–Microsoft. The Gartner Group says that the top

four OS opportunity areas for ISVs are Microsoft, Solaris, HP-UX and Linux. But

Microsoft has been playing its cards close to its chest, from the IA64 launch

onwards. While a few modules of Windows 2000 64-bit OS have been commercially

released, the whole approach has been pretty low key, and details about future

roll-outs and commitment to IPF are few. Itanium was the exception in Intel’s

golden rule-book, when it announced a product without Microsoft next to it on

the starting block. This time around HP replaced Microsoft on the block. So what’s

happening? Is this Microsoft’s cold shoulder, or is it too preoccupied with

the development of .NET services?

Gartner’s hardware analyst Boon believes Microsoft is committed to the

Itanium family rollout. "They’re not talking IPF, for full OS support is

not available as yet," he says.

The road ahead



The Aberdeen Group points out that the greatest challenge in the roll out of

the Itanium family is following a schedule, with the consensus of ISVs and OEMs.

Gartner’s Boon points out that HP may have pressured Intel into announcing the

Itanium roll out to meet HP’s product roadmap, leaving Microsoft on the wrong

foot. But neither HP nor Intel can afford to do this again. With the launch of

McKinley round the corner, Microsoft’s involvement is now a critical success

factor, and Redmond could be calling the shots on the road ahead.

Will the long-term scenario be much clearer? Unlikely. End users live in as

much ambiguity as IT vendors. Do vendors know when a customer will migrate from

Unix to Windows? No. Often, the customers themselves don’t know. End-user

migration trends to the Itanium platform will be more than unclear. One way to

help customers get over the learning hill is to help them build Itanium based

pilot systems for major new apps (see sidebar Migrating to IPF).

The future is also likely to be strewn with heart-burn, changing loyalties

and improved fortresses. Says Bhandari of Intel, "More and more Unix boxes

will get deployed on Itanium. Some will continue where they are as

niches, while others may have to make tough decisions". What about vendors

like Sun who have decided to continue in their "closed eco-systems"?

Will they face rapid erosion in marketshare because of Itanium shipments?

Not likely, says Gartner’s Boon. For the next four years, few CIOs will

need to reevaluate their purchase decision of Sun servers because of the IPF, he

points out. And that lets Sun off the hook, at least for the time being. But

Hewlett-Packard’s Peter Hall insists that it’s a two-horse race, "and

Sun is not in it".

That’s not the only ‘option’ to Itanium, though. The cheaper 32-bit

Intel systems will continue to dominate the volumes. "That 32-bit market

isn’t going away," says IBM’s Maloney. "As long as Intel keeps

shipping 32-bit chips we will keep making the servers". IBM was probably

the first to launch a server running the Xeon MP (multiprocessing) chips. In

fact its only Itanium product is an x-series (its Intel series) server, the

rack-mounted x380 for data centers.

But there is no turning back. The move towards 64-bit computing is as

inevitable as the one from 16-bit to 32, or 8 to 16. And the 64-bit leap could

be as dramatic as the previous ones, even though the big impact will be in the

server rooms, and not on the more visible desktop–in this decade.

Arun Shankar



The author has been executive editor of Dataquest. He continues to write on
business computing arun_shankar62@yahoo.co.in