How we tested IBM's HS22

Our test of IBM's HS22 focused on three areas: CPU performance and power consumption, 10GbE network performance, and solid state disk (SSD) performance and power consumption.

Thanks to Ixia Communications for the use of their Optixia XM2 chassis with ASM 10G Ethernet cards for all network oriented tests. Ixia's IxLoad was used to automate the test procedures. All I/O tests were run with the iozone open source I/O test tool.

For CPU performance and power consumption, our goal was to compare the new Nehalem processor with an older Intel Core based architecture processor. We computed the amount of power needed to max out the processor capacity using a php script executed through a Web server over Fedora 10 running on the blade.

We measure power in watts used to power the blades in the IBM h-chassis through its PDU Web interface. There are two power outputs to monitor based on the chassis configuration. Power used for chassis fans is ignored. We recorded these values once per minute. An average number of watts was computed over the span of measurement, which starts 1 minute after the test starts and ends 1 minute before the test ends.

We also measured the number of transactions per second to saturate the processor, create a metric of watts/(trans/sec) this gives a relative amount of energy per transaction and compares energy per transaction for all runs of the test.

Return to test.