It is with a little bit of a tear in my eye that I report the demise of Atlantis Computing. It was well-known that it was looking for a buyer. I noticed that Atlantis employee Jim Moyle had jumped ship to FSLogix; this piqued my interest, as he had been there for the majority of the company’s life. I have had a bit of a love-hate relationship with Atlantis. Its original ILIO product, although technically interesting, was in the beginning as stable as an isosceles triangle standing on its point. However, the company had weathered that storm, and its USX product line was rock solid. I could have made this post a postmortem piece, but I will leave that to others.
The demise of Atlantis Computing got me thinking about the HCI sector in general. There has been a lot of consolidation in that space in the recent twelve months or so. Nutanix finally IPOd, SimpliVity was bought out by HPE, and Cisco got a little cosier with Springpath and Pivot3.
HCI is a valid infrastructure model. You bring your storage back to the compute layer, removing your distributed SAN. For some workloads, but not all, this makes perfect sense. Nutanix has had early mover advantage and has benefited from having its business model ratified by VMware’s introduction of VSAN. You could even argue that it is successful in terms of customers, but it is still losing money hand over fist, quarter on quarter.
And herein lies the issue I have with HCI vendors. Why would I do this when I can roll my own for much cheaper from a single vendor, such as with VMware’s vSphere and VSAN or Microsoft’s 2016 Hyper-V and Storage Spaces Direct? The HCI vendors would argue that they have special sauce in their little black boxes that makes them taste better. However, in this throwaway commodity world that we live in, do people still want that boutique experience from a hardware vendor? HPE and Lenovo (ex-IBM) are finding their traditional server market stranglehold squeezed by white-box providers (HPE suffered a 14.3% drop in unit growth in Q1 2017 over Q1 2016; Lenovo had an even worse drop, of 27.3%). Admittedly, the others include Nutanix, but the vast majority of devices are legitimate white-box servers by vendors like Supermicro.
What does all this mean? I think the HCI bubble may have burst. Sure, we will see certain vendors moving units, perhaps even moving toward profitability, but I truly believe that Atlantis is not the last casualty of this consolidation period. That said, Atlantis’s demise could be a result of its fairly recent conversion to physical HCI with the introduction of its HyperScale units. Prior to this, it was a pure software play with first ILIO and later USX. Selling hardware units is difficult. In fact, as a go-to-market strategy, it was a difficult sale. It was a technical sale in an era of C-level briefings, and from the perspective of a C-level exec, it was quite a high-risk decision.
Nobody, including arguably the forebear of the converged/hyperconverged revolution, VCE, has truly been able to solve this issue for the enterprise. vBlock’s vxRail and vxRack never really sold in great numbers (units sold, not money spent). It will be interesting to see if Dell can do any better now that it wears the mantle.
Other Issues with Modern HCI
HCI vendors like to talk about web-scale, but what exactly is this? Well, for one thing, it is not is a four-node rebadged Supermicro FatTwin SLED unit. Real web-scale vendors do not grow in four-node FatTwin units. Facebook, Google, and Amazon grow in units of multiple racks in size. Another issue for the HCI vendors is that the low-hanging fruit has been picked, and HCI vendors are now having to compromise on their vision to gain more customers or to drive further inroads into an incumbent customer’s infrastructure. Storage-only nodes and compute-only nodes are in reality anathema to an HCI model, but Nutanix and SimpliVity offer one or the other. Those entities are hardly an HCI appliance. A rather unfortunate downside of this compromise on this model is that it causes a significant reduction in performance for those workloads that are effectively running remote from their storage. This is one of the main reasons why Nutanix acquired PernixData: it was to return to its concept of data as being as near to the compute as possible, or data locality.
Atlantis Tech will live on in a new entity called HiveIO. These are an early-stage NYC-based startup with interests in the VDI/DaaS space. This is arguably the best place for USX to live. Hopefully, it will not move forward with hyperscale. What is interesting is that HiveIO has morphed out of the finance industry, which incidentally, is where Atlantis had the majority of its customers. One thing of interest it the fact the Kevin McNamara, HiveIO’s CTO, was head of Research and Development for JP Morgan End User Compute, and its Chief Engineering Officer, Ofer Bezalel, is also ex-JP Morgan, the same company that incidentally was one of Atlantis Computing’s earliest customers and investors.
I had several good friends who were employed by Atlantis. I wish them well for their next position. It is their customers who I truly worry for, though.
Interesting perspective. You say “HCI is a valid infrastructure model” but ultimately work to dismiss it. And you dismiss it because of “webscale” and then denigrate without considering what it is the customer wants or define what you consider ‘webscale’ means or it’s value.
I help run a user group – many people don’t want ‘webscale’ .. they just want ‘scale’. I grant you there were a lot of players in the HCI .. specifically Software Defined Storage space, and now there are a smaller amount. But that’s what you naturally get for a startup space – yes?
Your real ask should be “what are small to medium size companies wanting to do with their tech, where will enterprise go?” A lot, still don’t trust cloud. For larger enterprises, definable modules is useful as they scale: they too are still aren’t fully trusting of Cloud.
You tell me what webscale “isn’t” – but not what it is. You don’t tell me why webscale is relevant.
Until you can you can do that – HCI will be relevant, regardless of the players.