Today?s large distributed Windows NT environments can have hundreds of servers and thousands of workstations. It takes a truly herculean management effort to ensure the availability of computing resources across the enterprise and guarantee acceptable service levels for the end user community. The servers in these large networks are often very high-end machines with clock speeds up to 500 MHz. Despite the extraordinary computing power available the leading cause of controllable performance degradation lies not with the CPU but with the disk. Advances in chip design have realized significant gains in performance. Disk performance has improved too, but not to the same extent. Remember that disk I/O is still an electro-mechanical process and it has some finite speed limitations. Couple this with the fact that most common business applications are I/O, not CPU intensive, and it becomes clear how I/O bottlenecks can occur, and dramatically reduce system performance and end user productivity.