In "Making the Case for Server Consolidation" (Windows 2000 Magazine, May 2000), Sean Daily identifies two current back-to-the-future trends: server consolidation and server-based computing. A point Daily doesn't stress is that these two trends are diametrically opposed.
Daily makes a convincing case for server consolidation, citing reports from several industry research groups (among them Gartner, Forrester, and Computer Economics) that suggest that fewer servers lead to lower total cost of ownership (TCO). On the surface, this point seems like a no-brainer. Having fewer servers means fewer servers to break, fewer servers to maintain and upgrade, a less complex network, and a simpler job of reorganization if the company merges with another company or is reshuffled internally.
However, the consolidation model doesn't apply to the server-based computing model in general or to the application service provider (ASP) model in particular. Rather than consolidating servers, ASPs and ASP suppliers ensure that they can support as many servers as possible and expand capacity as easily as possible—and for good reasons. When you run applications from a central location, computing resources at that central location are vital, and those resources aren't limited to CPU cycles and available physical memory. That is, if you experience a bottleneck at the network I/O level, adding another CPU doesn't necessarily help anything. Conversely, losing one server in an application farm impacts your user base much less if the loss represents one-twentieth of computing power, not one-tenth. And those who maintain server-based systems—especially those who maintain the servers for ASPs—must plan for excess capacity and future capacity. Adding servers gets you much more than adding memory. Particularly for those who use Windows NT to support their customers (and, these days, a respectable number of ASPs do), the upper limits for the number of CPUs and amount of memory that a server can support mean that server consolidation isn't practical.
Product and service offerings in the ASP space bear this out. I don't see ASPs looking for ways to reduce the number of servers they need; rather, they are finding ways to make many servers work together better and to add servers to the mix without reworking the model. So companies such as Gateway (among many others) are coming out with rack-mounted servers for the ASP market. ASPs such as TSCentral are developing application load-balancing for their software. For ASPs and private customers, firms such as Blackstone are building "compute farms"—logical groupings of computers running load-sharing software that lets them share CPU power and memory, even when the computers run different OSs.
In short, although there are reasons to consolidate servers, distributed computing is alive and well in server-based computing, particularly in the ASP space.