How I tested
I took all six arrays and ran them in the lab for several weeks, running performance tests under Linux and Windows using software initiators and iSCSI HBAs from Alacritech and QLogic. In addition, all the arrays were tested for failure situations, both at the disk and the controller level. The Linux tests were conducted on a Dell PowerEdge 2800 server with dual 3.6GHz Xeon EM64T CPUs and 4GB of RAM running Red Hat Advanced Server 4.0 using the beta channel Linux iSCSI initiator (Version 188.8.131.52-2) and the QLogic QLA4010 iSCSI HBA. The Windows tests were run on a Newisys 2100 server with dual Opteron 250 CPUs and 4GB of RAM running Windows Server 2003 and using the Microsoft iSCSI initiator and the Alacritech SES2002 dual-port iSCSI accelerator.
All switching was provided by a Dell PowerConnect 6024 layer 3 gigabit switch, and all connections were copper, except for the QLogic QLA4010 iSCSI HBA on the Linux server, which was an SFP (small form-factor pluggable) fiber connection. Jumbo frames were enabled on the switch and used whenever possible.
The published performance numbers are based on read/write tests at varying byte sizes to better simulate actual application behavior. As with any benchmark, these are synthetic results, and performance under an actual application such as Microsoft Exchange will vary significantly.