4KB Random Read And Write
To read more on our testing methodology visit How We Test Enterprise HDDs, which explains how to interpret our charts. We cover 4KB random performance measurements on page two, latency measurements on page five and power measurements on page six.
Our test pool consists of 7200-RPM HDDs with varying capacities. The HGST Ultrastar He10 is the lone SAS representative, which will lead to higher power consumption than its 10TB SATA counterparts. It is important to consider these factors due to the direct impact on dollar-per-GB and power metrics. We tested the He10 in single port mode, but both the SAS and SATA He10 models feature the same performance specifications.
Our test pool includes the 10TB Seagate Enterprise Capacity v6, the 10TB HGST Ultrastar He10, the WD Gold 8TB, the 8TB Seagate Enterprise Capacity v5 and the 8TB HGST Ultrastar He8. We tested in CentOS 7.2 (CFQ scheduler) with an Avago 9305-16e HBA. All of the drives in our test pool are designed for 8+ bay deployments in enterprise RAID, SAN, NAS and DAS environments.
The HGST He10 leads the 4K random write testing with an impressive average of 405 IOPS for the majority of the test, while the Seagate 10TB and 8TB model provide ~380 IOPS. The Seagate 10TB and the HGST He10 both suffer reduced performance at the highest queue depths, with the Seagate dipping at QD128 and the He10 following at QD256. The trailing performance is likely due to the caching algorithms, but we also observe that the other three drives in the test pool, which all utilize similar caching techniques, continue to offer strong performance. The drives will rarely surpass QD16 in a real-world environment, so the dip may not be relevant to the majority of use cases.
The HGST He8 and the WD Gold 8TB both feature an almost identical performance profile at 315 IOPS, which is the result of using the same technology. WD recently conceded that it had "given up share" in the 8TB market, and the large performance gap between the Seagate 8TB and the WD Gold and He8 might be the source of the market shift.
The Seagate 10TB features a somewhat scattered performance profile during our testing, but digging into the Latency-over-IOPS and QoS charts reveals that it offers a better profile as workload intensity increases than the He10, and the previous-generation Seagate 8TB also exhibits why it is exceedingly popular. The HGST He10 encounters some latency turbulence at high QD, but it provides the best QoS measurement at QD32.
12Gb/s SAS always incurs higher power consumption than the 6Gb/s SATA connection due to its design (higher speed, dual-ported), which gives the competing SATA drives an advantage over the He10 during our power measurements. The SATA He10 model consumes 2.7W less than the SAS model on average (according to HGST specifications).
The Seagate 10TB flaunts an impressive reduction in power consumption compared to its air-based 8TB predecessor, but the SATA He10 likely draws roughly a watt less than the Seagate 10TB does. In either case, the He10 is impressive because we note that it consumes nearly the same amount of power as the SATA He8 and WD Gold 8TB, which highlights that even with a SAS connection it delivers a much better power-per-TB measurement (due to 2TB of extra storage).
The Seagate leads the IOPS-per-Watt measurements easily and outpaces the 8TB contenders. The efficiency metrics would also be higher for the SATA He10, as well.
The HGST He10 takes the lead during the random read workload and nudges the Seagate 10TB aside with a peak speed of 252 IOPS at QD256. The Seagate is well within competitive range at 245 IOPS at QD256, but a closer examination of the results reveals that it has a performance advantage from QD to QD64. Nearline HDDs often reside in large RAID arrays or object storage pools, which spreads the workload over many drives, thus making low-QD performance especially relevant. The Seagate 10TB leads the entire span of light workloads, with its only challenger being its 8TB predecessor.
We usually do not include scatter results for random read HDD tests, as the difference between competing products is typically small, and the resultant charts are either very hard to interpret or simply incomprehensible. In this case, we do get the benefit of a closer examination of the WD Gold 8TB and the HGST He8. Both drives have the lowest performance from QD64 to QD256, which separates them from the pack. However, it would be hard to tell the two apart because they offer such a similar performance profile during the tests that the results are nearly carbon copies; that is, one drive overlays the other neatly in the charts. We know that WD based the drives on the same technology, but the test results during this workload might convince some that they are actually the same drive.
These similar results carry over to the latency over IOPS, QoS and histogram tests. These charts also reveal that the He10 offers the best performance under heavy load, while the Seagate competitors are better during light workloads.
The rubber meets the road in the mixed workload test. Storage vendors regale us with performance specifications for either pure read or write workloads, which are the easiest to master, but in reality, most workloads feature a mixture of read and write activity. This test includes both read (100/0) and write workloads (0/100) on the left and right sides of the chart, respectively. We mix in more write activity as the test progresses to subject the drives to more relevant and challenging mixed read/write workloads.
The Seagate exhibits its proclivity for random read workloads as it leads on the left side of the chart, but the He10 offers more performance from the 70/30 read/write workload to the 30/70 read/write mixture.
Both the He10 and the WD Gold 8TB experience a significant performance decline at 20/80 and 90/10 mixtures. We identified this tendency when HGST debuted its game-changing media caching with its 15K HDDs, but subsequent products, such as the 10K and 7,200-RPM models, eliminated the tendency. The issue returns, but in reality, it isn't much of an issue. It would be a stretch to refer to these heavy-write random workloads as a corner case, but it is going to be rare enough, especially in light of the dominance of sequential workloads in high capacity applications, that it is almost a non-factor.
We also note one of the few differentiators between the He8 and the WD Gold 8TB in this test. The He8 continues to offer a steady performance profile as we reach the heavy workload mixtures, but the WD Gold dips.
