In performance tests of QLogic Corp.s SANbox 2, eWeek Labs faced an interesting problem: how to test a product that came out before testing-tools vendors had released tools for it.
2G bps is a fairly high-performance rate to drive through a single port, let alone the 15 remaining ones in SANbox 2. In the IP switch world, tools such as Ixias Explorer have the ability to drive the ports of Gigabit Ethernet switches all the way up to wire speeds. Unfortunately, similar tools are not easily available for SAN (storage area network) equipment.
When we started our SAN setup, we expected help from testing/diagnostic tool vendors Finisar Corp. and Spirent Communications Inc.
Finisar didnt have 2G-bps traffic generators at the time of our tests. Spirent reps told us they had a 2G-bps tool ready, but when it came time to test, the company pulled out because it found bugs.
For throughput testing, we used Intel Corp.s venerable Iometer storage benchmark, far and away the most used test in the SAN market, but it was designed to pound on RAID subsystems.
Running Iometer in a standard configuration doing sequential reads across two RAID partitions on a Hitachi Data Systems Corp. Thunder 7200 RAID unit, we drove roughly 85MB per second through each of the partitions simultaneously. This was less than half the rate we wanted to push through the switch port. (Switch ports should max out closer to 200MB per second.)
To get faster data rates, we used a trick that we usually see at trade show exhibits. By tweaking the test specifications to use a very small section of our RAID partitions, the entire workload was small enough to be cached on the RAID controllers memory, which essentially eliminated the hard drives as a potential bottleneck.
Although this setup doesnt resemble a real-world configuration (data never hits or gets transferred from the hard drive), it allowed us to push close to 160MB per second through the switch ports, which is fairly close to our initial goal of 200MB per second of traffic.