This month at Hitachi NEXT, Hitachi Vantara unveiled their latest high-end storage platform, the VSP 5000 Series. With up to 21 million IOPS and latency as low as 70µs, the figures are impressive. But do we need another all-flash storage array?
Hitachi has a long history of producing very solid storage arrays for the enterprise. The VSP, USP and 9900 series have delivered high performance and exceptional reliability for many years. VSP remains one of the few systems to offer FICON (mainframe) support, as well as traditional Fibre Channel and iSCSI protocols.
There is an argument that these high-end storage arrays are perhaps somewhat over-engineered (more on that in a moment), however, from my experience when you’re deploying storage for applications that run banking applications which trade into the trillions of dollars, reliability is a pretty desirable factor.
The VSP 5000 series (of which there are currently four members) offers storage capacities from 23PB to 69PB internally (using 30TB SSDs), with the ability to virtualise other platforms for a total virtualised capacity of up to 287PB.
Hitachi is claiming 21 million IOPS with performance as low as 70µs and up to 149GB/s throughput (on the high end 5500H model). You can find a full set of specifications here. I would like to see more data on exactly how latency changes with throughput. Simple “as good as” figures provide little indication to customers on what to expect from the platform with their workload. This is an issue all vendors are guilty of, not just Hitachi.
It’s interesting to note that the platform supports NVMe and SAS SSDs, as well as the custom FMDs (Flash Module Drive). The FMDs have been around for many years and were a solution to offloading encryption, compression and otherwise compute-intensive functions into hardware. As a comparison, think DirectFlash from Pure Storage (similar, but not entirely the same). It seems that the FMDs are being “de-emphasised” as there is currently no NVMe FMD on offer.
Another interesting aspect of the configuration choices available is the continued use of SAS drives. Hitachi claims that the 70µs latency can be achieved with SAS-connected flash. That’s due in part to the new interconnect between nodes called “Hitachi Accelerated Fabric”. This is a PCIe switched interconnect that uses FPGA acceleration.
Hitachi Accelerated Fabric
I took some pictures of the Accelerated Fabric at Hitachi NEXT (see the image in this post). From the discussions I had, it appears the technology is 100% proprietary and used as a way to accelerated inter-node communications. Although the image show the controllers being fairly closely coupled (physically), the interconnects are optical. This means controllers could be spread further apart than a single rack, in order to extend resiliency. You can find out more on the Accelerated Fabric is this white paper.
Getting back to the question of over-engineering, it’s clear there is a lot of proprietary technology in VSP 5000. That’s what Hitachi does well – building solid hardware solutions. I don’t see this as a bad thing because the initial move here is to transform from using custom ASICs to programmable FPGAs, similar to those we’ve discussed recently.
It was interesting to note that Hitachi also demonstrated a software-defined implementation of SVOS, the VSP storage operating system. Looking at the design, I wonder if there is the possibility of seeing mid-range solutions that use commodity components such as SmartNICs and create high-performance solutions using commodity components.
In any case, there are already other vendors delivering all-flash storage solutions with proprietary hardware. The hardware/software-only debate will continue to swing back and forth as it has done for many years. There’s no right or wrong answer. Instead, customers should be looking at their requirements and determining what works best for them.
So, is the VSP 5000 simply Hitachi catching up with the market? The answer is probably yes – and no. Certainly, VSP 5000 brings Hitachi in line with the capabilities of other major storage vendors, even exceeding the performance capabilities with 70µs latency. Remember that this has been achieved without the use of storage-class memory and with SAS drives. So, it’s likely we will see more performance gains from the VSP 5000 platform. However, there’s currently no NVMe over Fabrics (NVMe-oF) support (yet), which may (or may not) be of interest to high-end enterprise customers.
The guarantee of eight 9’s of availability is certainly more than any other vendor has offered in a single platform. This translates to around 300 milliseconds of downtime in a single year. Remember of course that is isn’t an absolute number, but an average over all arrays in the field. Effectively, Hitachi is guaranteeing 100% availability, because even a small outage on one array would reduce the field percentage significantly. As with any guarantee, check the small print with your vendor.
The Architect’s View
I would expect nothing else than solid engineering from Hitachi and I’m sure the VSP 5000 series will live up to expectations. Enterprise customers that need extreme performance and availability will, no doubt, be happy to transition to the new platform. My only reservation here is how Hitachi addresses the breadth of market requirements. Modern hardware is now so reliable that many applications could survive on less-engineered solutions.
Hitachi VSP F and G-series arrays fit these requirements, and many could be delivered with SDS SVOS. I would like to understand how these solutions interoperate to allow data to migrate between platforms as necessary and to provide a single data plane and management solution. The value then becomes one of a fully-featured portfolio of products, rather than just having the fastest product on the market.
Copyright (c) 2019 Brookend Ltd. No reproduction in part or whole without permission. Post #82ad.
Disclaimer: Hitachi Vantara paid travel and accommodation for Chris to attend Hitachi NEXT 2019. There is no requirement to blog or produce content from the event. Hitachi has no editorial direction over any content produced.