SMB, iSCSI And Standard Server Workloads
Single-Client SMB Performance
QNAP's TS-451+ performed well in the synthetic measurements, so it's not surprising to see it leading many of our real-world tests over SMB.
These benchmarks are traditional home and small business NAS metrics, where users transfer files to and from the system. For most folks, this is exactly why you buy networked storage.
Multi-Client SMB Performance
Each node in the multi-client test falls short of saturating our test subjects. We used real Microsoft Office software to record traces that run from each system with a dedicated gigabit Ethernet connection to the network. The combined workload does tax the GbE link. With some products, we hit other hardware limits before banging on Ethernet's ceiling.
We can see this happen on two of the charts; either throughput sharply decreases or the latency spikes. I like to use latency to determine the load level at which a NAS device still delivers acceptable performance. As you can imagine, it takes a lot of traffic to make a NAS fail outright. But you don't want the nasty user experience preceding that failure, either.
Single-Client iSCSI Workloads
iSCSI is the most underused feature on NAS appliances today. All of our comparison products have the feature installed right out of the box, but we rarely hear about it being used. iSCSI adds a drive letter to the volume and shows up on your computer as a local disk. It's a great way to deal with large game libraries, particularly since the data on a RAID 5 array is redundant.
These results are presented in time units rather than throughput.
QNAP delivers strong iSCSI performance, giving its TS-451+ the advantage it needs to come out on top. The throughput result is about as fast as a 7200 RPM disk inside of your own desktop.
Since none of the systems scale as queue depths increase, the snake charts that show IOPS compared to latency are nearly vertical. Only a small number of users will deploy their low-cost storage appliances in one of these single-workload applications, though the systems are powerful enough.
Response time starts out around 8ms across the board. You never want to push much beyond 100ms of latency, and 200ms is a practical maximum.