As flash storage moves to mainstream adoption, a new breed of products are set to become the new tier 0 storage. However with the demise of DSSD, is the market set to falter?
Tier 0 has come to represent the fastest of the fast, almost a specialist category of super fast storage. The tier 0 mantle was previously held by the first set of all-flash storage vendors, such as Violin Memory and Fusion-io, both of which no longer exist. Today, we think of tier 0 as being defined by DSSD, Mangstor and E8 Storage, with performance capabilities in the hundreds of microseconds and millions of IOPS. These are custom hardware platforms that take advantage of new technologies like NVMe, Infiniband and directly attached PCIe. If you need I/O performance in an external shared array, the new tier 0 is the way to go. So why has DSSD been discontinued and why are firms like Mangstor going through a revamp? There are a few issues:
Market Size – The market size for tier 0 is undoubtedly small. The Register speculated in December of last year that Dell EMC had sold less than a dozen DSSD D5 arrays for around $6m in revenue. Consider this in terms of the spend on DSSD – around $1.3bn, including acquisition and ongoing development. With that rate of revenue generation, getting your billion back is going to take some time. The original flash market also had a potentially small market size, but this quickly changed, as vendors started to use commodity flash components to either ameliorate their existing architectures or to build new. The storage market was ready (and we could say desperate) for flash because of the cost and relatively low performance of HDD systems, meaning there was a latent demand. It’s difficult to see the same demand for new tier 0, although there are definitely use cases.
Architecture – All-flash systems came off the back of custom hardware designs. Look back at Kaminario and you can see a platform that evolved from a DRAM appliance, through to a NAND/DRAM hybrid with Fusion-io cards to today’s commodity based offering using TLC flash. Vendors that were stuck with custom flash hardware (e.g. Violin) haven’t survived, unless of course you’re at the high-end like Hitachi or IBM and have big pockets. The rest of the market has moved towards commodity and had to compete with a raft of software vendors offering all-flash hyper-converged and scale-out solutions. Just look at the marketing dollars that have gone into VMware Virtual SAN and you see what I mean. Although they claimed it wasn’t (when I directly asked the question at TFDx), Violin definitely had an issue keeping up with flash development. In such a competitive market, custom hardware doesn’t offer the agility needed to compete in a crowded arena. Architecture also has to include features; today a rich set of space optimisation and data protection functionality is essential.
Application Capability – How many applications are currently capable of using such as high performance platform? This issue is related to market size – without a range of suitable applications, the market isn’t going to grow.
So is the new tier 0 over before it’s begun? Not really. The vendors who are more likely to be successful are focusing on software-based solutions, rather than hardware platforms. Recently Excelero announced their NVMesh architecture, based on commodity hardware – servers, NVMe drives and Ethernet NICs. The result is a scale-out node-based storage solution that aggregates capacity from multiple servers and NVMe drives into a single “mesh” architecture. Storage is presented as block-based LUNs to servers in the configuration. You can get more details on the technology by watching their presentations at the most recent Storage Field Day event. E8 also uses commodity components, as does Mangstor.
The Architect’s View
- Excelero presents at Storage Field Day 12 (Tech Field Day Website, published 8 March 2017)
- DSSD President quits Dell EMC (The Register, published 9 December 2016)
Comments are always welcome; please read our Comments Policy first. If you have any related links of interest, please feel free to add them as a comment for consideration.
Copyright (c) 2009-2017 – Chris M Evans, first published on https://blog.architecting.it, do not reproduce without permission.