About the only cache tuning action that can safely be conducted on most Windows 2000 machines is monitoring cache size and cache effectiveness, and adding memory to machines displaying signs of diminished cache effectiveness. In this section, we examine a series of file caching experiments that compare and contrast effective versus ineffective Windows 2000 file caching. These experiments were conducted using the Windows 2000 Resource Kit’s Performance Probe program for generating artificial file I/O workloads. In the first scenario, we show how effective the Windows 2000 file cache can be when the cache size is adequate. In the second scenario, we observe the Windows 2000 Cache Manager under stress, attempting to manage a file access pattern that resists caching. Finally, we examine the impact of the one Cache Manager tuning parameter that is available for controlling the size of the Windows 2000 file cache. All tests were conducted originally on a Windows NT Server running Version 3.51 on a Pentium Pro 200 machine with 64 MB of RAM installed. Subsequently, we verified that the identical behavior occurs under both Windows NT 4.0 and Windows 2000. The tests were conducted on a standalone machine. The only other application running concurrently was Perfmon, which was logging performance monitor data to disk during the test runs.
Scenario 1: Effective Caching
In the first scenario, we ran the Probe program, available in the Windows NT Resource Kit, defining a 16 MB file ...