Fewer servers, better service

Think big. That may be the image ingrained in our imaginations, but it’s not always accurate when it comes to managing IT resources.

One big idea currently percolating at Dresdner Bank AG focuses on thinking small. The Frankfurt, Germany-based bank is in the process of paring down the approximately 5,000 servers it maintains throughout Europe.

Depending on what’s running at an individual branch, network administrators must manage a wide variety of server platforms, including IBM Corp., Hewlett-Packard Co. and Sun Microsystems Inc. hardware running each vendor’s flavour of Unix or Windows 2000. This IT potpourri evolved as the headquarters and branches added bits and pieces during the past decade, according to Anthony Wilbert, systems manager at Dresdner’s IT services arm. “It’s cheaper to have one big machine than hundreds of small machines,” he said.

The bank’s consolidation project began about a year ago and will continue indefinitely. So far, the main result has been the installation of a Sun Enterprise 10000 server at headquarters. The bank’s goal is to reduce its total number of servers, but Wilbert acknowledges that some branches will continue to run local applications on local servers.

Better IT . . . With Reservations

Another “think small” advocate is Wyndham International Inc. in Dallas. The global hotel chain is putting the application that controls guest arrivals, departures and billing onto two IBM eServer p680 computers running Unix. The same guest-management tasks used to require 165 servers scattered across 100 locations.

The Web helped push Wyndham to re-evaluate the role that today’s mega-boxes play in a wide-ranging enterprise. A few years ago, the hotel chain picked Opera, a thin-client application from Columbia, Md.-based Micros Systems Inc., as its guest-management system because the program allowed IT managers to upgrade and tune the program at all of its front desks from a central administrative centre in Dallas. “We then wondered, ‘If that’s good at the workstation level, wouldn’t it be good at the server level, too,’ ” recalled Gary Owen, vice-president of IT operations.

Although companies expect to save costs through server consolidation Wyndham estimates it will lower its hardware bills by as much as 40 per cent in the coming year learning to live with less comes with stumbling blocks. Managing fewer servers may be easier, but when things go wrong, everyone suffers – not just the workers at a single network node. Likewise, network scalability, security and availability become so critical to a firm’s health that there’s almost no room for error.

Wyndham and Dresdner Bank aren’t the only large companies pruning the tentacles of their client/server networks. Ten years ago, the placement of servers everywhere and anywhere was considered the best way to achieve high performance and computing availability. Nowadays, corporations are taking advantage of better networking gear, lower costs for high-end components like switches and administrative tools that manage a small number of servers in central locations.

Server consolidation doesn’t necessarily mean that large enterprises are less hungry for hardware, only that their tastes are changing.

Companies start down the consolidation trail to reduce costs. Besides lower hardware costs, Wyndham expects to save “several hundred per cent” over five years in reduced travel and training costs, according to a Wyndham spokesperson.

Consolidation can also make networks more manageable. Programs like HP’s venerable OpenView, Computer Associates International Inc.’s Unicenter TNG or Tivoli Systems Inc.’s TM give administrators centralized control panels to keep servers and networks running at optimal speeds by monitoring performance and, in some cases, reallocating resources to break through bottlenecks.

Reducing the number of servers that a company runs also helps minimize the demand for scarce IT expertise. Instead of needing multiple systems managers, companies can operate with a handful of administrators at IT headquarters. New generations of networking hardware and services have created price/performance advantages that make consolidation possible.

“We now see sophisticated technologies, like [symmetrical multiprocessing], even in very inexpensive servers,” said Jonathan Eunice, principal analyst in the server technologies group at Illuminata Inc., a technology research firm in Nashua, N.H. “The result is excellent performance in a two-way server for about US$2,000, or a fully configured eight-way server for under US$100,000. This makes it technically appealing to do a consolidation.”

He added that a new mind-set is taking over server management techniques. Rather than buying high-end computing power packed into towers, large companies are putting together racks of servers that fill a central data centre much like the telecommunications gear that hums in communications closets. “By consolidating 50 one-processor servers into new-generation server racks, companies save not just on equipment costs but on the costs of power, management and maintenance contracts,” Eunice said.

Finding a Balance

Although large companies are finding that server consolidation offers hope for lower costs and easier systems maintenance, IT experts caution against blindly tearing out the old and plugging in the new.

First, a company considering consolidation should look for a balance between a widely distributed and consolidated approach. “Either can be right. There’s never a single correct solution,” said Mark Hudson, a spokesperson for HP’s 9000 servers. Consolidation efforts should be driven by individual applications, he said.

“There will be areas where a distributed environment makes sense, especially when a specialized application is needed locally,” Hudson said.

Companies with a large number of affiliated branches, such as Dresdner Bank, are often better served in a mixed environment. “Even today, a central data centre is probably not good enough for, say, 800 offices,” Eunice said. “You may really need a server in each branch. Here, distributed makes all the sense in the world. On the other hand, if you’re dealing with a large office, [as in an] insurance company, [where] the network within a single building or campus is so reliable, so close to 100 per cent available, [then you] don’t really need servers on every floor, and you can have a single data centre.”

Companies can also take advantage of less-is-more computing by consolidating at the local level. For example, a multibranch company might use centralized servers for individual campuses, buildings or floors. Savings in hardware and maintenance costs will still accrue, if on a smaller scale.

“The real answer is to be as centralized as the situation allows,” Eunice said.

The crucial consideration, no matter how widely a company embraces consolidation, is system reliability. If all the computing resources are put in one basket, managers must watch that basket closely.

“Really understand what service levels you want to provide,” Hudson said. “Up front design is critical.”