Who Needs 64-bit Computing?

Software Magazine, Oct, 1998 by Alan Earls

You can count companies like federal mortgage lender Fannie Mae and others with high-performance needs among the early adopters. But not everyone's ready to scrap their tried-and-true 32-bit technology.

If a picture could tell a story, Kodak IT Manager Gus Holder's most representative snapshot would show his company's highly automated warehousing operation. The image is complex, realtime, and mission-critical. The operation stocks 120,000 discrete types of products at five major locations, and operates one-and-a-half miles of sorters and conveyers and five miles of autonomous guided vehicle lanes. But that picture could still only convey a small measure of the complexity. Beyond the shop floor, with its 350 barcode readers, there are 1,500 desktops tied together and linked with Kodak's ERP system and a massive data warehouse.

Indeed, for most IT managers it's hard to fully appreciate the challenges Holder's IT operation faces or even its sheer size: 58 CPUs, 5.5Gb of memory, and 2Tb of DASD. Sitting at the top of the heap is a pair of IBM SP2 mainframes, which, among other things, run a data warehouse that adds 350Mb of new data nightly. Since it was put in place, it has maintained detailed, instantly accessible records for all operational transactions over the past two years.

It's an IT infrastructure of which Holder is mighty proud. But it is also far too vulnerable. A natural or man-made disaster in Rochester, N.Y., where his data center is housed -- or even a short-lived IT glitch -- could shut down operations across North America in an instant, and cause financial losses to mount rapidly.

The solution, of course, is a disaster recovery plan and backup. But providing functional duplication of the key SP2s would mean incurring tremendous costs. So when IBM offered an alternative -- its new 64-bit S70 -- it took almost no time to make Holder a buyer.

"Economics was the key driver," he says. "We brought in an S70 on a trial basis, and the loading of the apps was so much faster that it was obvious, even without tuning, the performance was there."

Without bothering to do a detailed analysis, Holder concluded that a single S70 server would be sufficient to back up several SP2s and, down the road, would be able to scale well beyond current plans to increase capacity by installing even more SP2s in the future.

Now, the S70 hums away in a remote, unattended operation near Toronto, ready to shoulder the burden if necessary.

But while Holder clearly feels a 64-bit machine was the right solution for this particular problem, he, like other IT managers, is not yet ready to jump ship on tried-and-true 32-bit technology. More to the point, however, there are still precious few applications written to take full advantage of 64-bit addressing.

Indeed, as Dan Kusnetzky, program director for operating environments and serverware services at International Data Corp., observes, "it is almost as if vendors are rushing to 64-bit only to get there before some other vendor does, not because a mass of end users demanded it."

For his part, Holder is counting on IBM's plans to continue to expand the capabilities of the SP2 architecture. For now, though, he's content knowing that 64-bit can deliver substantially higher performance, even with code that is not 64-bit native. And for his "worst case-only" app, that's still a good enough fit.

Remortgaging The Future

While Holder has chosen to limit his use of 64-bit computing, a small cadre of early adopters has taken the next step -- investing in 64-bit Digital AlphaServers as a cornerstone of future operations. One of the more high-profile examples is Fannie Mae. Established in 1938 under the appellation Federal National Home Mortgage -- Fannie Mae is a private corporation, federally chartered to provide financial products and services that increase the availability and affordability of housing for low-, moderate-, and middle-income Americans.

By one measure, Fannie Mae has the distinction of being the largest corporation in the U.S., with $366 billion in assets and an additional $674 billion in mortgage-backed securities outstanding. The company is one of the largest issuers of debt after the U.S. Treasury and its stock is traded on the NYSE.

What makes Fannie Mae successful is, in part, its ability to accurately predict which mortgagees will be reliable debtors and which will be likely to default. "Whoever analyzes the risk best can price best," says Eric Rosenblatt, senior credit analyst, thereby increasing the size and productivity of the loan portfolio. "On the margins you will end up much less relaxed about taking a deal when your equations are producing less specific results," he says.

The problem is not unlike that described by Coleridge's Ancient Mariner -- there's water everywhere but nothing to drink. Fannie Mae has more than 10 million loans, each of which comes with complex transactional records. But slogging through that sea of data takes lots of processing horsepower. The problem is aggravated by the type of processing that needs to be done, where the questions are nearly always open-ended and the answers are difficult to predict.