In the first part of this chapter, we conclude our examination of server and server-related architecture by systematically revisiting the major conclusions of this work and then considering possible evolutionary paths. The first part complements the conclusions of the first three chapters, which performed a similar role for server-relevant technologies; here, we look at how these technology projections could be exploited in possible future systems architectures.
The second part of this chapter is a general conclusion.
Looking at the semiconductor technology trends, in which increased densities allow the integration of multiple systems elements processors, caches, controllers, and so on onto a single chip, it is clear that low-end and mid-range servers (say four- to eight-processor SMPs) will become commodity items. As a result, only vendors with huge volumes will be able to continue to play in these markets. Although multiprocessor workstations are emerging, we should note that server-level requirements for reliability, availability, and maintainability mean that workstations will remain very different from servers.
As we have emphasized several times, the major interesting aspect of clusters is their high intrinsic availability. The key problem with clusters is the lack of a standard programming model to make parallel programming straightforward. Although UNIX cluster products are functionally very similar, they all build on different interfaces to provide that functionality, in particular continuity of service and performance growth. Windows, having just one maker and just one set of interfaces, has no such problem.
Clusters fit naturally into three-tier architectures,...
TABLE OF CONTENTS 
