Violin Memory, one of the early high flyers in the All Flash Array (AFA) space, filed for Chapter 11 bankruptcy in December 2016. This blog discusses some of the issues around their predicament, and takes a look at how the AFA market's use of custom flash modules (CFMs) (which Violin used in the Flash Storage Platform) has been impacted in enterprise-class arrays over the last couple of years.
The demise of Violin Memory, one of the early high flyers of the AFA market, has not been greatly exaggerated. The NYSE suspended trading in Violin Memory shares on October 28, 2016, and delisted its stock because it had not maintained an average global market capitalization of $15M over 30 consecutive trading days. From a revenue high of around $108M for their fiscal 2014, the company steadily shrank revenues in the wake of their September 2013 IPO. The company officially filed for Chapter 11 bankruptcy in December 2016, and an unsuccessful attempt to auction off company assets in late January 2017 has left company employees and customers a bit in the lurch.
Several things accounted for Violin Memory’s trajectory, all of which have been commented on in the press over the last year. The company did not evolve quickly enough as AFA market requirements shifted, and a maturing market, populated with strong AFA offerings from all of the established enterprise storage providers, made it difficult for the remaining smaller players, particularly when those vendors were not growing but were shrinking year-over-year revenues like Violin. In IDC’s Worldwide All-Flash Array Market Shares, 3Q16: Top Five Market Share Holders Consolidate Gains (IDC, March 2017), IDC noted that the AFA market, while still the highest growth external enterprise storage market, has reached a point where there is broad feature parity between the top 6 market shareholders, and issues like a company’s go-to-market strategy and ability to sell effectively will be the determinants of future market success (rather than differentiating features). While Violin’s latest Flash Storage Platform had a lot of strong features, as a company they clearly did not have the ability to match the go-to-market strengths of competitors like Dell/EMC, NetApp, HPE, Pure Storage, and IBM.
Violin was one of several vendors that chose to go with CFMs instead of commodity off-the-shelf (COTS) solid state disks (SSDs) for their array. I had the opportunity to publish a CIS document last year that discussed the differences between CFMs and SSDs in enterprise-class AFAs (Flash Media Packaging Decisions Secondary to System-Level Considerations in the AFA Market (IDC, September 2016)). Over the years, several vendors had chosen to use CFMs instead of SSDs (EMC, HDS, IBM, Pure Storage, Skyera, Western Digital) in at least one of their AFA products. Custom hardware is often used when target market requirements can’t be reliably met with COTS components, and the vendors that chose to use CFMs made strong cases for why they thought their choice gave their customers better performance, endurance, reliability, and density with lower energy costs and at a lower $/GB packaging cost than COTS SSDs would have in the same time frame. The inexorable march of commodity product towards “bigger, better, faster” over time that is a hallmark of high tech is tough to beat, however. As COTS SSDs increased in performance, endurance, and density while costs continued to drop, three of the vendors’ CFM-based AFAs have basically left the market. Skyera was absorbed by HGST (who stopped selling the AFA), and HGST was then later acquired by Western Digital. Violin Memory’s Chapter 11 bankruptcy filing signaled their impending market exit, and just this month EMC announced that they would no longer be selling the EMC DSSD D5, a rack scale flash AFA that used CFMs. And the availability of the Samsung 16TB SSD, now used by NetApp and HPE in their enterprise-class AFAs, have removed the storage density advantage that CFMs offered from all vendors except Pure Storage (who uses a 52TB CFM in their new FlashBlade AFA).
In doing research for the Flash Media Packaging Decisions document, I encountered very strong opinions from HPE, NetApp and Kaminario that the development of custom hardware is at best risky in evolving markets. A quick review of the use of CFMs over the course of AFA market evolution indicates that early on, a high percentage of industry revenue was being generated by CFM-based products, but that that contribution has steadily dropped, and for shipping products today, only HDS, IBM, Pure Storage and Western Digital have AFAs using them. Revenues for all of these products (HDS VSP F Series, IBM FlashSystem 900/V9000 and A9000/R, Pure Storage FlashBlade, and the Western Digital InfiniFlash (and the IBM DeepFlash 150 Elastic Storage Server, which is a product based on the InfiniFlash hardware)) are on the upswing, but their revenue growth is strongly dominated by large all-flash platforms that are growing in triple digits (EMC VMAX All Flash, HPE 3PAR StoreServ 8450/20450/20850), NetApp All Flash FAS) that use SSDs.
Interestingly, both the Pure Storage FlashBlade and the Western Digital InfiniFlash products are not targeted at the primary flash space where the other products are targeted – they are going after an emerging secondary flash market IDC calls “big data flash”. Big data flash is for use with very large, unstructured data sets where there is a strong requirement to support high degrees of concurrency, and requirements for storage density, $/GB cost and flash throughput and bandwidth are more important than low latency, and there is still a strong case to be made for the use of custom hardware in this market. In fact, to date none of the products targeted at the big data flash market (there are only four shipping for revenue) use unmodified COTS SSDs. Nimbus Data is the fourth vendor shipping a product in the big data flash space (ExaFlash), and IBM (DeepFlash), Pure Storage (FlashBlade), and Western Digital (InfiniFlash) are the others. That may change in the future, but this is a much younger market than primary flash and there is more of a runway for custom hardware to generate revenues based on relevant differentiating functionality before commodity offerings catch up.
As the pre-announced 32TB and 64TB SSDs begin to ship in early 2018, we may see big data flash revenues start to shift in the direction of COTS SSD-based systems instead of CFM-based systems, but at this point it is too early to tell. As vendors build larger systems designed to simultaneously deliver on low latency and high throughput and bandwidth, there are potential concerns with COTS SSDs (those are gist for another mill, possibly a future blog). But there is no doubt in the primary flash space that revenues are increasingly shifting towards COTS SSD-based systems, and Violin’s demise over the last 18 months has been a contributor to that. Their choice to use CFMs did not drive their demise, but it is clear that the trend is away from AFA revenues overall being generated by CFM rather than SSD-based AFAs.