10
Windows speeds ranged from 19MB/s to 78MB/s with an average of 42MB/s when using
the HBA. The top speed is slightly less and the average is the same compared to the
software initiator results.
Analysis
There was no significant performance difference between the software initiator and the
hardware initiator. This means that iSCSI (over gigE) could realistically be rolled out
without any hardware cost for the hosts. Additionally, the simplicity of the setup on each
host translates to minimal cost in terms of time required by administrators to setup the
hosts.
On the storage side, the only additional hardware was an iSCSI router for existing FC
attached storage. If new storage will be purchased as part of an iSCSI solution, finding
disk that natively supports iSCSI would eliminate this cost as well.
However, the cost savings above come at the penalty of performance. Compared to other
storage architectures (FC, native SCSI, etc.), iSCSI is relatively slow. At approximately
40MB/s, regardless of initiator type, it is most useful for adding more space on single
machines. It could also be useful for copying or migrating data over a long distance
when only tcp/ip networks are available, but existing tools like
rsync and scp already
serve this purpose. This is all based on using 1Gb/s Ethernet for the transport, which is
the limiting performance factor. Assuming a stable network and an existing investment
in 10Gb/s and faster networks, the performance may increase enough to make iSCSI
useful for other applications.
