The server's memory and L2 cache (processor cache) usually do not become bottlenecks because they are not service centers; that is, processes don't wait for these resources. If there is insufficient memory in the system, the system starts paging. This increases the visit rate on the disk, and the disk becomes the bottleneck. This situation is commonly known as disk thrashing. When you investigate a disk bottleneck, you must also examine the paging rate. If the paging rate is more than 20 pages per second, consider adding memory. Although memory isn't technically the bottleneck, adding memory can solve the problem by relieving the disk.
An inadequate L2 cache can cause the processors to stall while waiting for data from main memory. Excessive cache misses can cause the system bus to become a bottleneck. This is commonly known as bus thrashing, and is an element to watch for, especially in multiprocessor systems. If you believe this is happening in your server, you can install the Pentium counters available in the Microsoft Windows NT Resource Kit compact disc and monitor bus utilization by using Performance Monitor.