Beyond the predisposition of departmental managers to maintain control of their computing assets, the politics of decentralized computing gets even more contentious. The threat of re-centralization is even more offensive to systems administrators than it is to traditional business managers, because this type of power play can have a detrimental impact on their careers. While many systems administrators may split their time between user support and systems management, the possibility of relinquishing control of systems in the name of efficiency also carries with it the likelihood of diminished technical responsibilities. The rational departmental systems manager will therefore fight vociferously to maintain these responsibilities. I know I would.

There are other political barriers to server consolidation. In many organizations, all central administrative support organizations, from human resources to purchasing to IT, are viewed with a certain level of contempt, since these organizations typically act in ways that constrain departmental autonomy. Certainly, the central IT organization makes decisions every day that inhibit departmental flexibility in choosing technology, often without prior consultation, and the resentment engendered is only intensified by the fact that market forces have driven salaries and annual raises considerably higher within the glass house than outside it.

But It's Not Just Politics While politics is often the most powerful force working at odds with consolidation efforts, technical impediments also exist. Some are fairly creative. I've heard many departmental managers, and even a few central IT folks, argue against server consolidation because a single server's failure will have less of a negative effect if there are more servers. Carried to its logical extreme, this supports the argument for a single server for every client, or maybe even two servers each if fault tolerance is considered, a strategy that would surely increase overall systems reliability. This strategy of distributing risk is largely the invention of inept IT managers who either haven't figured out how to provide reliable system services or who are simply looking for an excuse not to ruffle the feathers of departmental business managers.

Clearly, there are valid technical obstacles that stand in the way of consolidation, and the most significant is the lack of adequate network bandwidth. Consolidating servers requires a fundamental rethinking of the old 80-20 rule that drove backbone designs of the past. When LAN services were decentralized, it was reasonable to expect 80 percent of the bandwidth utilization to be local to any given LAN segment. However, when basic LAN services are centralized, the impact on backbone resources can be dramatic. In many cases, the 80-20 rule is still valid, except that only 20 percent of the traffic is local to a particular segment while 80 percent traverses the backbone.

While this was once a potentially insurmountable obstacle for campus or high-rise LAN environments, today it's a simple exercise in network engineering. The cost of switched Ethernet has fallen below $50 per port, and with similar economics driving Fast Ethernet and high-speed routers, coupled with responsible deployment of VLANs (virtual LANs), it's becoming increasingly cost-effective to design hybrid switched/routed LAN infrastructures that provide plenty of bandwidth to support centralized server farms. That's not to say this kind of overhaul is easy, but as earlier-generation shared-media hubs approach the end of their lives, it pays to build an infrastructure that takes traditional packet congestion out of the equation.

Such a strategy isn't possible when you're supporting users across a geographically dispersed network that includes low-speed remote connections. In this case, you either need to adapt to a distributed server-management strategy or take a more radical approach using thin-client systems.

Even if LAN capacity and performance are not problems, server scalability may be a serious issue. From a hardware standpoint, the picture is rapidly improving, with server processing power following price-performance improvement curves similar to what we see on the desktop. Beyond raw processing power, improvements in I/O subsystems are also noteworthy, with faster network interfaces and dramatically improved cached fault-tolerant disk subsystems now available in the commodity marketplace. At the system software level, leading Unix vendors have made significant strides in setting performance standards for database applications, and Novell continues to build on its legendary file-server performance, pushing the envelope to 1,000 concurrent users on well-configured servers.

Unfortunately, organizations that have sought to replace Unix and NetWare with Windows NT servers have discovered that they pay a significant cost in performance and scalability. These costs may be offset by reductions in expenditures associated with integration of file, print and application services on a single platform, but I remain highly skeptical in that regard. Long-term advancements in clustering technology may help to eliminate some of the scalability concerns of all these platforms, but don't count on functional and stable implementations before late 1999.

Beware False Conclusions Don't misread my words as a blanket endorsement of consolidation or an indication that consolidation will never happen. Instead, think seriously about the potential benefits as well as the political and technical obstacles that stand before you. Also, don't assume that consolidation strategies are always black and white. It may be possible to consolidate some services while delivering others in a cooperative arrangement that meets both central IT and departmental interests. In my next column, I'll outline some strategies for success in both areas.

Dave Molta is director of network and system services at Syracuse University. He can be reached at dmolta@nwc.com.