Windows 2000 Performance Guide

Benchmark Testing

Configuration of disk arrays is almost never a straightforward process because there are so many options and parameters to consider. The most important decisions are whether to use hardware RAID or software RAID, which RAID organization to use, and the length of the stripe unit. It is important to understand the various trade-offs in performance, reliability, and cost that result.

EMC CLARiiON

In this section, we describe a number of experiments we conducted on the EMC CLARiiON disk array under Windows NT to explore the effects of the many configuration options. Note that these tests were run using a Windows NT server as opposed to a Windows 2000 server. Although the results may not be completely representative of what to expect with a Windows 2000 server, the exercise of running these tests and the amount of analysis we add to the results should be well worth the effort of reading through them.

Benchmarking methodology

The experiments evaluate the throughput of an external storage processor configured for RAID Level 0, RAID Level 5, and RAID Level 0/1 logical disks. To explore the advantages and disadvantages of each organization in detail, we applied three types of workloads:

• A workload of only reads

• A workload of only writes

• A mixed workload with a 65:35 read:write ratio.

A matrix showing the scope of our experiments is shown in Figure 10-24.

Figure 10-24. Test ...