Not me. Henry Newman at Enterprise Storage Forum. See article here.
His first prediction of more consolidation in the SSD space is a given. I’ve been arguing that for a while. On the fab side, there are what … four producers left? Toshiba/Sandisk, Samsung, Intel/Micron, Hynix? Did I miss anyone? Will any of them leave (voluntarily or otherwise)? I think the SSD space that will really consolidate is on the SSD-as-a-rack-appliance side, as well as on the card side. Despite massive raises, the “why” question sorely needs to be answered, as well as “is this a viable business”. I think the market has spoken pretty clearly on that at this point.
Nothing on his second prediction, I agree. It will slip.
His third, that FC is going the way of the dodo, is obvious. Its been obvious for a while. I am not being critical of Henry here. Its taken years for people to get this (again, not pointing to Henry on this, he’s been a truth teller on this for a long time). But this has long term implications for “SAN” technology. We’ll get to that.
Prediction four, big data appliances … yeah, dead on. We are doing that. It was an obvious step for us, as we were building very fast tightly coupled systems for a long time, and we started spending more time in the application space helping customers tune their installs. A fairly large group of customers in finance love our boxen. Everyone of them has asked us for a set of appliances to tackle specific issues. You can see the first wave at the link above. What’s interesting is when prospective customers test out their applications in PoC’s on our appliances. Phrases such as “order(s) of magnitude faster” are bandied about.
This one is a no brainer for us, but highly problematic for non-purpose built systems. Take a unit that runs very efficiently and very quickly, and you can replace many competitive systems with a single system of much greater performance. Then you win on TCO, acquisition cost … everything.
We’ve got more appliances coming out. One might argue, well, gee, won’t the public cloud kill this business? Much like OCP killed servers? Since the latter didn’t happen … well … no, I can’t reason that way. Public cloud is being adopted, of this there is no doubt. But private cloud growth is bigger, faster, more compelling. But whats really interesting is that the private AND the public clouds benefit from this. One of our larger customers has us building their public cloud for them. Out of the same appliance bits we build private clouds from. High efficiency matters, high performance matters. High performance == lower costs.
Henry’s fifth prediction is interesting, one we are getting many requests for. One we are working on some technology for as well.
His sixth prediction, that someone will actually do something about storage security … I have my doubts, in that some of the fundamental models are broken. I think we need encryption for data at rest as well as in transit for one, but we also need sane, reasonable, and above all else, simple to use access control for data. The poster child for this epic failure is likely to be the ACA in 2014. It used to be the law of the land (HiPPA) that people could go to jail for improperly sharing private health information about patients. Something tells me that rather than fix the data access and storage issues (not to mention all the other issues), the rules around this will be relaxed. So, I think I disagree with Henry on this, though this is an area we are actively thinking about. I know Veristorm, and a few others are looking into this for big data analytics. But you need this in the storage side as well.
Henry’s seventh prediction on tape … long, long ago, I bought into the mythos around tape, and advised my old research group to get an exabyte tape backup system. It was about a $20k purchase in the early/mid 90s. They used it for a while, then the drive died. They couldn’t get a replacement drive. What struck me at that point was that a backup technology was only as good as the drive. The medium itself, if unreadable due to dead drives, was irrelevant. At that point, in the late 90s, I became a fan of putting backups on multiple disks. I’ve not changed my mind since then, and seeing technology continue to change, and vendors and their drives changing/disappearing, I think we collectively need to rethink long term archival. Badly need to rethink it. Tape isn’t it. I’d like to believe that drives are better, but long term bit rot and mechanical issues are problematic. How do you store data, and make it accessible for 100 years? For 1000 years? How much research and development data will be lost when media die or become unreadable (same thing)?
His eighth prediction is dead on. Our big data appliances include parallel file system appliances specifically to tackle this market.
I do not agree with Henry’s ninth prediction. If anything, I see something completely the opposite. I see SAS being the new FC, and people looking to alternative technology within units. SATA itself is fine for some use cases, and these use cases will continue to expand. No one really cares if a technology is “enterprise connectivity” if their storage and application have built in intelligence to handle errors correctly. More to the point, I see this intelligence being pushed up the stack, and storage being made somewhat dumber. Which plays well into the SATA model.
FWIW, I’ve heard that “SATA is dead” for many years now, not unlike “this year is the year of 10GbE” and “Infiniband shall die this year”. None of these have ever come to pass. I won’t argue whether or not they should come to pass, just simply noting that these haven’t and likely won’t.
Henry’s tenth prediction may come closer to truth, though I expect that given the performance of object storage and erasure coding, that people will start to abandon massive scale RAIDs at some point. We can build single RAID units at 1/3 PB size now in a single box. The issue with RAID is fundamentally the block stripe read and write which is terrible for random IO apps, not to mention the rebuild times. I look at declustered RAID as a stepping stone to leaving RAID behind at some point in favor of a POSIX layer atop an object store (Ceph for example). This will take time, and I don’t expect it to happen in a year. 4-5 maybe, so declustered RAID will have some life to it.
Note: I’ve been notoriously wrong about any number of things in the past, including betting my early career on GaAs. Because, you know, its much faster than Silicon, so of course it will win.
More to the point, I highly, strongly urge people who make predictions to read carefully Arthur C. Clarke’s famous story, Superiority. In the intro I read for it, it was once required reading at MIT for a specific course on industrial design. Its all about how being better (SAS, etc.) can cause you to lose the war to inferior designs (SATA, etc.)
This said, I hope Henry is right on most of these. And we are betting our company on a few of them.