I was very interested to see the latest IDC data on external storage arrays* and shocked to learn that spinning disk is not actually dead yet. In fact, the market for Hybrid Flash Arrays (HFA) was still larger than All Flash Arrays (AFA), at least for this quarter. So, what gives, why do spinning rust arrays still rate a mention at all?
The current situation reminds me of my bemusement when I arrived at Hitachi a few years ago. I came from a networking background, not a storage specialist by any means, and I naively asked why we even need a category of “AFA” when you could obviously plug disks or flash into these boxes. If you want an “AFA” just configure it with all flash, why was this even a category?
“Who on earth hires networks guys?” Was the most common answer.
I remember there was much hand wringing among the sales team that we could not get a mention on the AFA quadrants because our VSP at the time could be configured by choice to be flash, disk or both. To be eligible for the AFA quadrant, you needed to be only and exclusively configurable with Flash, and one needed to be on that latest and coolest quadrant to keep up with the new cool kids.
Well, as a non-storage guy at the time this had me totally confused. To my mind this would be like creating an “All 100G” router category which would exclude routers that could support multiple interface speeds through pluggable optics. To a network engineer, I would have found it ridiculous, and I’m sure my customers would have too. So it was that I found an “AFA” category that excluded arrays capable of Flash or disk or both, quite perplexing.
It seemed to me that the AFA category had the effect of marginalizing vendors that had the ability to offer both in the same box, for no good reason at all. “Quadrant Proliferation” I would have called it at the time. Well, turned out I was “educated” that actually there are architectural benefits to excluding spinning disk and optimizing for Flash and it was on this basis that the separate category should exist. So significant would these architectural differences be that AFA would blow an all-flash configured HFA off the map – obviously!
Well, I guess I could wear that, though in the same way that I would expect the same argument to apply for “100G only” routers, I would have expected this architectural benefit to show up in real world performance and not require the arbitrary protection scheme of a distinct quadrant.
Turns out the real-world performance was not the quantum leap that a separate quadrant would imply. In fact, many vendors made the transition to the AFA quadrant through “SKU-Engineering” and made a variant of their HFAs available in All Flash only configurations. Guess what, they got on the quadrant and kept up fine with the AFA cool kids on the performance front.
So, was the alleged performance benefit of restricting the media choice for the sake of a quadrant category really a win for customers? It certainly was a win for new vendors not having to support “legacy” media types but for many customers the tradeoff of cost vs performance for spinning disk was still worth it. After all, if you are willing to trade off performance and cost to tier to “the cloud” why would a middle tier of cheaper on-premise storage not also be a choice you might want to make?
Of course, Flash is faster, better and getting cheaper and will obviously “win” the storage wars but not cheaper enough in less performance-sensitive uses to kill the disk just yet. It may be shocking to some that HFA is still outselling AFA (at least this week) but it seems that choices rather than quadrants are still calling the shots.