4-Corner Performance Testing
Sequential Data Transfer by Size
Regardless of all the extra features, the TS-453mini is still an NAS, and eventually you will want to read and write data to it. In the benchmark results below, we compared the QNAP TS-453mini against four other four- and five-bay systems. The prices vary among all of the systems, with some costing more than the TS-453mini and some costing less. All of the devices were armed with 4TB Seagate NAS hard disk drives. RAID 5 was also used across the field of products.
In all but one test, we use one gigabit-Ethernet port back to a Netgear S3300-52X 100/1000/10000 switch. We use a Quanta MESOS CB220 server on the other end to administer the workloads. The MESOS CB220 connects to the network via 10GbE Twinax.
Before we jump into the real-world workloads, we must first evaluate the systems with synthetic tests. This helps us understand where a product is strong and where manufacturers need to optimize performance.
The QNAP TS-453mini delivers consistent sequential performance across the block size range. We didn't observe any wild dips when reading or writing. As you can see in the charts, many of the systems deliver around the same sequential performance. The ASUSTOR AS7004T does have a leg up on the other products in some of the write tests, but it's also the most expensive model, costing nearly twice as much as the lowest-priced system in the charts today. The QNAP TS-453mini sits in the middle of the price range.
Random-Data Transfer by Size
Delivering consistent sequential performance is easy compared to the explosive nature of random data. Most of the data you transfer to your NAS is sequential in nature. Movies, music and pictures — the staples of NAS storage — are all sequential. As these systems became more powerful, they moved closer to the PC. In some configurations, they are the PC. Applications read and write a lot of random data. If you use a NAS as a secondary drive that holds programs that you run over a mapped drive or iSCSI, you are using a lot of random data.
We should always see a drop in the random charts as the performance lines move from the left (very small files) to the right (larger files). The TS-453mini performs as expected in both random tests but isn't the highest-performing across the board. The TS-453mini does give users the option of using an SSD for random-read cache. Users can also use an SSD to cache sequential data, but this isn't an optimal use for the flash, a resource that costs considerably more than hard disk drives.
In an office environment, the NAS is often accessed by many users at the same time. This increases the stress on the system. Each new transfer increases the queue depth. Again, we get nice smooth performance from the TS-453mini that is comparable with the other systems in the comparison.
Writing data at varying queue depths is more difficult for many NAS appliances. When we get to high queue-depths, the sequential data becomes random because we're pulling small pieces of data from the disks and from different locations on the platters.
The ASUSTOR AS7004T dominates this test, and there is a reasonable explanation for that. The AS7004T uses a Core i3 desktop processor and not an Intel SoC low-power chip like the other systems do. The Celeron J1900 in the TS-453mini is a powerful processor, but it isn't Core-class.
Some of the systems in the charts turned in unpredictable write performance in our synthetic tests. This comes from the cache in the systems where data is buffered in DRAM and the amount of time the system was subjected to the workload. We're working to contain the wild shifts, and our extended tests seem to have done the trick. This will be the last review with the older models tested using the older write tests.
Sequential Mixed-Data Sweep
The data-sweep tests show mixed read-write IO performance. We measure two block sizes for sequential and two block sizes for random transfers. Here, we see that a lot can change between two traditionally tested sequential block sizes in the 70 percent read mix. Smaller files put more load on the system and require more IOPS to complete. The TS-453mini compares well against the other products in the chart.
Random Mixed-Data Sweep
The random sweeps run just like the sequential sweeps but here use 4KB and 8KB blocks. The 4KB blocks make up the majority of Windows read-and-writes when the applications are held on the NAS and accessed over iSCSI. VMware virtual machines are tuned for 8KB random blocks.
The TS-453mini gives us another solid performance with increasing IOPS as we add more random data writes. As we add more writes to the mix, the internal cache aids in increasing the IOPS until we get to 100 percent writes with heavy cache.