Sun storage system delivers high performance
ISCSI-based 7410 unified storage system has some usability and integration issuesFollow @infoworld
Cluster configuration at first glance was very easy, a simple matter of letting the system detect the second connected Sun Storage 7410 and telling it to add the second system to the cluster. The redundant controllers can be set up in an active-active or active-passive mode. Active-active provides two separate storage pools, each with its own IP address. If one of the controllers fails, that controller's pool is taken over and served by the other controller. In an active-passive configuration, only one controller is active, serving one storage pool. If the active controller fails, the passive controller takes over. The active-passive controller is less complex to set up, and has a faster switch-over time in the event of failure, while the active-active system has less utilization under normal circumstances, and provides two storage pools rather than the one provided with an active-passive configuration. Failover takes a little over a minute in active-passive mode and about 30 seconds longer than that in active-active mode. In either case, the iSCSI initiators on the test servers lost the connection and had to be manually re-connected to the iSCSI volumes.
While we could test failover features of the cluster, testing the controllers for performance in a clustered configuration was not possible because of an unresolved issue with the system Sun sent to be tested, symptoms for which included freezing of the administrative interface, spurious reports of drive failures and failure of the ILOM interface.
Therefore, we pulled one of the controllers out of the test bed and proceeded with all performance tests on a single controller, seeing as the clustered configuration doesn't add any extra performance considerations. It would have been our preference to set up each of the four available iSCSI ports on this single controller as a separate port on the same subnet of our network, but this configuration is not supported by Sun at this time (although Sun did tell us it is working on this kind of support for a future release). The Sun Storage 7410 requires a different network for each port on the controller, which could be up to 16 ports if you used all available slots for four-port gigabit cards. This is a clumsy and inefficient way to set up storage because if you need to change the servers around for any reason, and each one has a different subnet configuration, managing the pool of servers becomes more difficult.
Following instructions from the Sun engineers, we then set up all four ports as a single aggregated connection using link aggregation control protocol. However, because the control interface uses one of these same four ports, we had to designate one port as an admin port, and three as aggregated iSCSI ports.
Performance of the single controller system, as far as our limited test bed could verify, was excellent.
The controller managed an average of 67MBps throughput per gigabit connection. We did not have enough servers to generate enough traffic to max out the aggregated connection. However, taking our base numbers and extrapolating them out to a scenario where there were 16G Ethernet connections, the number would come in around 1,072, which is very close to Sun's throughput claims for the Sun Storage 7410. That assumes you didn't encounter any scalability issues along the way, of course.
With four gigabit Ethernet connections across our IOmeter-driven tests, we were unable to move CPU utilization on the Sun Storage 7410 system above 3% with the average of 1,600 IO/sec each of our four connections, which bodes well for the system's ability to support the 30 or 40 servers necessary to generate 288,000 IO/sec maximum Sun advertises.