REPORT ANALYSIS AND FINAL THOUGHTS
The Kingston DCP1000 is very good at delivering more IOPS per QD in most workloads due to its multi-volume design, however, there are drawbacks to this. While it is a sequential workload and read workload monster and delivers great performance in mixed and write workloads, it isn’t as consistent in performance during those workloads leaving there much to be desired.
With many of the big players in the industry focusing on consistency, it is bit unusual to see the Kingston DCP1000’s performance falling short under these workloads. Reflecting on that and the provided specifications of the Kingston DCP1000, it is actually hard to hold that aspect of its performance against it. After all, the Kingston DCP1000 enterprise SSD is a read-oriented data center enterprise SSD from the get-go. Kingston and Liqid built the DCP with this specific task in mind and knew that there might be a give and take when pushing for the highest performance of an SSD in the industry. It is designed for and will be best utilized in more read oriented applications or applications that, while needing the high performance it has to offer, don’t need such endurance.
Another observation of the multi-volume design shown with the DCP1000 is that it consumes more power and produces more heat. It is essentially four SSDs mounted to a single PCIe board. While it is rated for 35W consumption while active, it drew 40W from our power module during specific workloads as shown. We might suggest that, if you plan on deploying these cards in your servers or server solutions, airflow requirements of 500LFM at 35C or 600LFM at 40C should be met.
Lastly, one might also consider that with the design of the DCP1000 having four controllers, a bridge chip and the SMDs, there are more cogs in the wheel that might cause failure. From another view however, this could be considered a good thing for failover if you have it set up in a mirrored RAID in software or are utilizing it in software defined storage applications, just as it was designed for.
With all that said, our views at days end are that the Kingston DCP1000 is a total beast. It is an IOPS spewing monster that, when utilized to its full potential, can deliver more than enough performance for the vast majority of adopters. Move over Intel DC P3608, we have a new performance king! In today’s testing, it was able to sustain sequential read and write speeds of 7GB/s and 6GB/s, respectively! Not only that, but it is able to deliver over 1.1million IOPS with 4KB random read performance and over 180K for write. During the server workloads, it beat out most of the test pool, except for the Micron 9100 MAX during the more write heavy workloads at higher QDs. From lower and mid-range QD’s the Kingston delivered similar if not better performance though. Of course, it was able to beat out every other drive we have tested in our read oriented web server workload where it left over a 150K IOPS gap between it and the Intel DC P3608.
Looking back at our SNIA inspired graphs, while the IOPS and average latencies looked great, we couldn’t help but notice high maximum latency results that were very concerning. Diving deeper into our results of that test and our other tests, it seems that it is only related to that specific workload. During our other tests, we saw maximum latencies that were similar to the other SSDs in the test pool comparisons. After reaching out to Kingston they as well did some more testing and replicated the result, but only in Windows. Utilizing Linux and FIO rather than Windows and Iometer, they did not observe these drastically high maximum latencies. Furthermore, they have reached out to some of their customers real-world testing the DCP1000 and thus far, none have reported similar high latencies in their POC testing or after the drives had been deployed into their production applications. Still, we might expect to see a firmware update somewhere in the near future. So, from this, end users shouldn’t see similar spikes in their workloads.
Finally, looking into its efficiency, we can say that while it does consume a good amount of power, the performance it spits out can easily make up for it. What other SSDs can deliver over 7GB/s read and 6GB/s write bandwidth and over 1.1million read IOPS?
The Kingston DCP1000 wouldn’t be alive if it wasn’t for the partnership between Kingston and Liqid. Over the past few years, Liqid has been leveraging Kingston’s engineering and manufacturing expertise to bring the DCP1000 to life. Liqid provided the turnkey design to Kingston for the DCP1000, meaning they take care of the controller aggregation over PCIe switching, provide the code that lives on the ARM processor, tuned the M.2 FW for latency and endurance, and tested and validated the design. As well, they provide all the manufacturing test tools and test flow to the Kingston factory. Kingston then takes all this and throws in their experience to enable the mass production of these SSDs, open up distribution channels, and deliver exceptional customer support. Together they have created the current performance king and to make things even more interesting, they are now working on a U.2 PCI NVMe version as well. Let’s hope they continue down this road to help bring some more high-performance SSDs to market and help to boost the performance of our everyday lives.
It goes without speaking that the Kingston DCP1000 Enterprise SSD is deserving of our Innovation Award, as Kingston may have just redrawn that performance line in the sand. Absolutely beautiful product and highly recommended!