Toward A More Flexible FutureToward A More Flexible Future

Early business adopters of blade-server technology will use it for simple activities. But if blades prove their worth on the edge of the network, running firewalls and E-mail, more companies will consolidate back-end servers onto a modular chassis, running databases and vital transaction processing. As a result, look for IBM and Intel to push development of more powerful four-way Xeon MP and Itanium 2 blades next year, and for competitors such as Dell and HP to follow suit.

For blades and bricks to carve out a place in the world of enterprise computing, they need to take market share from proven Intel-based server technology that still appeals to many companies and is delivering the cost savings managers want. In April of last year, Menasha Corp., a $1.1 billion-a-year printing and manufacturing company, moved its SAP system from RISC-based HP 9000 servers to Intel-based Dell PowerEdge servers. The move has saved the company about $650,000 in hardware and ongoing maintenance costs, CIO Edward Wojciechowski says, because Intel servers are less expensive to buy and maintain.

Menasha also offers an example of how the line between high-end and low-end servers will blur during the coming years. Wojciechowski believes he can get the power and flexibility the company needs by clustering Intel-based servers and avoiding a single point of failure. "I'm not looking at high-end servers today," he says. "I like to have smaller servers for more specific apps. I like this idea of buy what I need."

Intel's plans for the coming years match that thinking: produce chips that continually push the edge of what lower-priced, low-end servers can handle. HP has committed to standardizing on Intel Itanium chips, which it co-designed, and phasing out its RISC architecture and Compaq's Alpha chip by 2006, with support through 2011. But it doesn't want to lose customers or Compaq's reputation for highly reliable servers. So HP plans to port its HP-UX version of Unix to the Intel chip, and by 2004 to transplant Compaq's Open VMS operating system and clustering technology to HP-UX. By middecade, HP expects to offer one standard, low-cost server that runs a high-powered version of Unix as well as Windows or Linux. Longer term, the company could even move the superreliable NonStop Kernel operating system, which Compaq acquired when it bought Tandem Computers, to Itanium.

Unlike HP, Dell hasn't fully embraced the Itanium 2 platform. When Intel released Itanium 2 in July, Dell execs said the 64-bit processor hadn't yet displayed the performance benchmarks and compatibility with business software to fit Dell's mass-market computing strategy. Last month, however, the company reversed its position, saying it would likely ship specialized Itanium 2 servers in coming months to run high-performance clusters. Russ Holt, VP and general manager of Dell's enterprise systems group, says Dell is keeping its eye on the Itanium 2 market even though it chose not to be first to market with Itanium 2-based servers. The limited success of the first version of Itanium is a big reason Dell will be more cautious with Itanium 2, Holt says. IBM plans to offer an Itanium 2 version of its Intel-based xSeries servers during the fourth quarter.

In the low-end server market--four or fewer processors--Sun sees the greatest demand in two-way servers. As two-way systems become faster and offer greater capacity, they'll increasingly be used to consolidate single-processor systems and replace larger servers, says Laura Finklestein, group marketing manager for Sun's low-end servers. Sun's response is a combination of hardware and software. In addition to the LX50 two-way, Intel-based server revealed in August, Sun is touting its upcoming N1 software and services strategy to manage distributed servers, storage, and networking devices. By the end of this year, Sun will launch the first piece of N1--virtualization software that centralizes data-center management by turning each system into a virtual LAN. By mid-2003, Sun will offer the second N1 component, provisioning software to automate the configuration of the virtual LANs. And by 2005, Sun is promising policy-based automation software that lets users set business policies that dictate how the system allocates resources. As computer systems expand, the test of their value will be how well companies manage data flowing through servers and other system components, says Chris Kruell, a manager with Sun's enterprise systems products.

High-end and midrange servers will also grow in speed and power. As with their low-end counterparts, much of the innovation for servers with eight or more processors will come through software and management tools. These tools enable clustering and partitioning, allowing companies to get more out of their server investments. Within the next year, IBM will extend its Cluster 1600, introduced in August to allow easier management of clustered servers, so users can dynamically reconfigure partitions for its p690 and p670 Unix servers.

Innovation is also reaching the mainframe, which remains the technology of choice for some. The ability to consolidate a large number of servers and applications onto one machine, with well-established management tools and a long record of reliability, has great appeal to some companies. IBM signed 75 new mainframe customers last year and expects to top that number this year. "Regardless of modular computing or the clustering of low-end Intel-based servers, there is still a place for mainframes, and that's not going to change anytime soon," Galileo's Wiseman says. Galileo replaced part of its mainframe infrastructure in May with SunFire 6800 servers to run its airfare-pricing system, but it still depends on mainframes for its transaction-processing systems. Wiseman says mainframes have a bad reputation because they're seen as old, proprietary technology. "But they work quickly and they work well," he says.

More importantly, they're learning new tricks. Advances in Web services, logical partitioning, and performance redistribution let mainframes run newer operating systems and applications. Mainframes from IBM also run open-source operating systems such as Linux. IBM is developing technology to let IT managers dynamically allocate and reallocate workloads on the mainframe. The goal is simplified management and increased flexibility of IT resources. This lets companies run online processing during the day, for example, and switch resources to maximize the efficiency of batch processing at night--all without reboots. Unisys Corp. offers this with the ClearPath Plus Server, Libra Model 180, introduced last month, which can be partitioned to run seven operating systems on a single server. Performance redistribution is just one of the features that users of high-end servers can look forward to. HP and Unisys are working on improvements to their capacity-on-demand features that let users add processing capacity in minutes and subtract that capacity just as quickly.

What gives mainframes the prospect for even greater longevity is the potential to use Web services. By using Web-services standards such as the Simple Object Access Protocol and XML, the cost of integrating a customer's travel-booking system with Galileo's services will be reduced by at least 75% in most cases, Galileo CIO Mickey Lutz says. Plus, building an Internet booking engine using Galileo's global distribution services will require less technical knowledge, Lutz says. Before the incorporation of Web services into Galileo's architecture, travel agencies needed in-depth knowledge of Galileo's systems to effectively integrate Galileo applications with their own.

The real value of a server boils down to its ability to efficiently run a company's applications, says Robert Hendricks, VP and CIO of Atlantic Health System Inc. Atlantic Health, a network of hospitals serving nearly 5 million people, relies on HP rp8400, rp7400, and rp5400 servers running HP-UX for its Horizon Clinicals apps from McKesson Corp. While vendors are making the choice among low-cost servers and high-end mainframes and everything in between less distinct, Hendricks sorts through the clutter by focusing on a few simple rules. "Whatever I buy, it has to be scalable in terms of transaction throughput," he says. "And I'm looking for a reliable track record from both the product and company."--with Paul Travis

Illustration by Randy Lyhus