The easiest place to begin evaluating NFS read and write throughput is on a single client and a single process basis. While measuring in a controlled and isolated laboratory situation is advantageous, it may not be practical. Measuring the actual environment of interest can provide more meaningful information as well as establishing a methodology for monitoring performance in the future. Under such uncontrollable situations, it is important to determine range and normal variation in measurements with particular attention to peak and non-peak periods of activity.

Once the NFS environment has been selected for measurement, some simple techniques can be used to initially evaluate NFS read and write throughput. One of the easiest methods is to use commands available on the NFS client system, such as copy or cp, to generate NFS reads and writes. Copy commands along with date, time, or timex commands can be used to determine the elapsed time for the copy. Throughput can be readily calculated from the elapsed time and the size of the file.

Some additional considerations when measuring NFS throughput are the effects of graphical user interfaces (GUI), any overhead associated with opening files, and I/O to and from local physical storage. Measuring with and without a GUI will show the effect on NFS throughput caused by the GUI. Using a network analyzer can assist with isolating the time spent opening files from the time spent executing NFS reads or writes. Techniques such as copying to /dev/null can eliminate local physical I/O for NFS reads. For more complex requirements such as reading and writing at random offsets, you may use an existing application to provide such access or write a program to execute NFS procedure calls at random offsets within a file. Program generated NFS procedure calls can also be useful when available local physical storage is insufficient for the file sizes to be measured.