Comparing Filesystem Performance
The case study presented here investigates the difference in the performance of disks based on the filesystem they are formatted with. To compare them, we simply create a number of volumes of varying sizes, format them with a filesystem, and generate a workload against them. The first question that arises is what kind of workload would bring out the differences between the filesystems. Your first reaction might be to use the Iometer tool from the previous chapter but unfortunately, it is not suitable for this test. Iometer either generates a workload against a physical disk or against a single file on a disk partition. It is meant to be used to compare the hardware components of the I/O subsystem, so it does not necessarily put the filesystem code to work.
We chose to use the Postmark benchmark developed by Network Appliance. It is available from the web site for free and it includes the source code (http://www.netapp.com/tech_library/3022.html). It was developed primarily for Unix servers, but since it’s written using standard C, it is very portable and compiles cleanly under the Win32 environment.
We decided to develop yet another benchmark because there is a lack of benchmarks that focus on typical application server workloads. Most servers currently provide services such as email, network news, and other applications that tend to deal with ephemeral files of small sizes. Postmark simulates heavy small-file loads for files that live ...