A revolution in servers is here

There’s a revolution going on in servers – one that’s easy to miss because it’s starting not in the enterprise data centre but in telecom switching, medical imaging and industrial electronics.

Once, servers was just another name for midrange or mainframe computers, with the major architectural variations relating to an operating system and to the number or processors in a system.

That’s beginning to change. Servers are migrating to more open or standard operating systems – such as Windows or Linux – and to more specialized use by their functions. Branded general-purpose servers first gave way to servers with preloaded software, then to hardware optimized for workload with the preloaded software. This is giving way to hardware and software that’s customized for a particular workload or application, also known as “appliance servers.” As if to ratify this migration, last year we saw an explosion in “rack-optimized” servers or systems such as the Sun Microsystems Inc. Netra 220R and 420R and the Compaq Computer Corp. ProLiant DL. In the first quarter of last year, rack-optimized servers accounted for 15 per cent of all entry-level server shipments. That doubled by the third quarter.

But that was just the beginning. This year, we’ll also see the following:

– Shipments of servers based on the InfiniBand architecture, which is designed to replace Peripheral Component Interconnect (PCI) buses.

– New enhanced workload management software for dynamically changing a server’s configuration or workload.

– New low-power processors that enable server rack populations in the hundreds and innovations building on these advances.

– Server “blades,” or motherboard computers that can be put into subracks and aggregated into larger systems, which are based today on the cPCI, or compact PCI, architecture.

Picture this: Instead of a single server in your data centre, you have racks of servers configured from numerous server blades, each with its own memory, processor and connection. Each blade can be dynamically loaded with software optimized for the function of the blade – for example, handling e-mail, Internet security or content caching. In the morning, you may need six blades to handle your e-mail, but only two after-hours, so the other four can be released for other functions.

Blades are already in use in telecom, military, medical imaging and industrial applications. Vendors include Force Computers, an arm of Solectron Corp.; Motorola Inc.’s Computer Group; and Ziatech, which Intel Corp. recently bought. International Data Corp.’s server analysts expect these vendors to expand their product lines first into the Internet infrastructure space – selling to Internet service providers (ISP) that support at least 100,000 end users – and then, as software improves, into enterprise data centres.

As these new technologies wend their way from the large networks and ISPs and into the rank-and-file data centres, what can you expect?

First, you can expect a mismatch in the hardware capabilities of racks of server blades and the software you need to manage and reconfigure them dynamically. You may be writing code yourself, but then again, you’ll have much noticeable improvement in your ability to guarantee performance to your end users.

Second, you’ll need some new skills in tech support, maintenance and systems management as well as design.

Finally, if you can’t adapt to these new technologies within, say, the next five years, you may find your data centre outsourced.

These new technologies will offer those who use them flexibility and performance over today’s general-purpose servers that are about equal to the advantages nuclear submarines have over diesel boats. Better dive in.

Gantz is a senior vice-president at IDC in Framingham, Mass. Contact him at jgantz@idc.com.