Sign in with
Sign up | Sign in

4K Results & Conclusion

Seagate vs. WD: Battle for the SOHO NAS Crown
By
Brought to you by What's this

4 KB Random Reads

Windows defaults to using a 4K cluster size. We’ve used this test for a long time as a way of sensing the “native” throughput of a storage solution when working with data purely optimized for Windows. Of course, this doesn’t reflect real world usage, but it serves as an interesting hypothetical.

Seagate clearly shows a 20% to 25% advantage across the board. Between the higher spin rate and workload optimization, the NAS HDD clearly earns its keep here. As a point of interest, compare the total throughput rate here (~ 0.6 MB/s) under a harsh, 100% random load against a pure streaming load (below).

4 KB Random Writes


Reversing the load once more, we dive into 100% random 4K writes.

The strangeness of this chart had to be triple-checked, but the results stayed consistent. With such high IOPS, it’s clear that the drives and/or RAID enclosure are performing caching – which is an excellent feature of the subsystem for helping to accelerate performance in certain conditions – but that still left the mystery of why we’re seeing almost perfectly inverse results from the two drives.

Our best guess here is that once WD queues this workload, the Red isn’t able to maintain the same caching efficiencies it did at a queue depth of 1 for whatever reason. With some workloads containing mixed reads and writes, writes can terminate the reads by looking at data prematurely, resulting in more cache misses than you might normally get. This could explain behavior like WD’s performance in this instance. Meanwhile, Seagate’s graph shows more benefit from read data and caching. While it’s hard to know exactly what’s happening without having engineers run bus traces on how the access patterns are being handled, it’s possible that WD’s drive has a bug manifesting in this specific work case. Given all the prior results, we expected to see very similar results between the two drives, so it makes sense that this particular workload is hitting a weakness.

4 MB Workloads by Threads

Finally, we need to examine how each drive model performs when its RAID gets loaded up with multiple simultaneous user requests. We can simulate this in Iometer by adding virtual worker iterations. Again thinking of a media-heavy use case, we opted for streams of 4 MB traffic files.

We end up with two clear bands here: streams based on 0% random traffic (higher throughput) and those with 100% random traffic (lower). In general, both drives exhibit very little difference in performance as user counts scale upward. Western Digital does show an interesting fall-off in the move from six to eight workers when writing is involved; the 100% read/0% random line does not exhibit this decline. Seagate’s drives do not mirror this decline. While the benefit is subtle, we can see that the NAS HDDs offer slightly higher performance for random traffic and more reliable performance with streaming 4 MB traffic as user counts increase.

Conclusion

Making grand statements here is difficult because the range of competition is so narrow. These are the only two drive models on today’s market that specifically target the 1- to 5-bay NAS segment. The Seagate NAS HDD units we tested, with their latest NASWorks firmware, do have the clear upper hand throughout nearly all of our tests. That advantage can range from nil up to roughly 30%, depending on the workload scenario. But when you consider that Seagate’s option is a bit more affordable, the gap between them opens even wider.

To be clear, these drives exist in a space between consumer-class desktop drives and higher-end enterprise drives, such as the WD Se and Seagate Terascale and Constellation family. It’s a bit unfair of us to throw enterprise-caliber tests at these sorts of NAS drives, but this is war, after all, and buyers would do better to see their drives under exceptional strain than something more appropriate to stand-alone drives, such as PCMark.

Speaking of which, we’ll be back soon with a different look at these NAS drives on a stand-alone basis. Running in a Synology enclosure under a RAID 5 configuration introduces one set of parameters on performance, but not all buyers may want such a setup. Perhaps they want to run a pair of these drives in, say, an internal mirror, taking the high capacity of desktop drives with the higher reliability of this NAS class. We’ll be very interested to see the next set of results and check for any changes in the story.

For now, Seagate’s NAS HDD emerges as our clear winner in the 1- to 5-bay SOHO/SMB storage race.