Four-Corner Performance Testing
Sequential Read Performance
Although Asustor products typically perform well in our tests, we ran into an issue with the AS6204T right out of the gate. We took our results to company representatives and had a new firmware one week later. The initial release suffered low sequential write performance at low queue depths. Asustor tells us the workload was so light at a queue depth of one that the system didn't enable Turbo Boost. Instead, it remained in a low-power state. The update fixed the issue, and is now publicly available.
Reading 128KB blocks happens a little slower than on some competing platforms, which is surprising for Asustor. Sequential benchmarks run using different block sizes look better, at least.
Sequential Write Performance
The sequential write tests are kinder to Asustor, though again the AS6204T encounters lower performance when it's faced with certain block sizes compared to other NAS appliances. Though the difference is small, we'll be keeping an eye out for correlation during the real-world tests.
Sequential Mixed Workload Sweeps
These results show how mixing reads and writes affects transfer performance. In a home environment, you might occasionally send and receive files at the same time. That happens a lot more often in an office. It's important that you buy the appliance that best matches your demands, and our mixed workload benchmarks help guide your decision.
Random Read Performance
You'll typically only encounter small blocks of data accessed randomly when you run software that uses the NAS as central storage. Examples include virtual environments like Hyper-V and ESXi (the former is optimized for 4KB blocks, while the latter is optimized for 8KB blocks). Other popular office applications, such as QuickBooks, use 4KB when run from a Windows environment.
Random Write Performance
Many of the random write tests look like waves as we move across queue depths. This is caused by the system cache when it flushes during our benchmark. We've increased the test duration, but the only way to completely remove those peaks and valleys is running each NAS in steady state. Although that'd help mask the cache's impact, we'd also veer further from reality. After all, NAS systems aren't subjected to the same taxing workloads encountered in the enterprise.
Asustor's products have always employed a lot of cache to improve random performance. NAS manufacturers can optimize this parameter from within the embedded Linux operating system. The AS6204T with 4GB of RAM doesn't excel in these charts as much as models that ship with 8GB of system memory, though.
Random Mixed Workload Sweeps
The effect of caching is obvious when you look at the sweeps across our random mixed workloads. When we hit 100 percent writes, performance spikes as the RAM absorbs incoming data. That steady build-up you see from 100 percent read to 100 percent writes happens when the cache steps in the help the slower hard drives.
In addition to our usual benchmarks, we also wanted to test with encrypted folders to see how Intel's Braswell-based SoC works with AES-NI active. In the tests above, we compare iSCSI and SMB transfers to SMB with encryption enabled.
Using encryption still imparts a performance penalty, but it's smaller here than on appliances without AES-NI. Previously, we saw encryption cut performance in half (or worse). That no longer has to be the case.