Why I'm Bearish On Fusion-io

It seems the whole world is talking about Fusion-io with the same reverence as it did Data Domain--and we all know how well that one ended up.

So why do I have the audacity to question what to many is a fait accompli already? Because I've seen this play before, Mrs. Lincoln.

Now, to be fair, the concerns I am about to raise (which are entirely correct, mind you) may have already been considered and addressed by Fusion-io, but if it has, no one told me about it. Thus, even if i'm wrong, I'm right. I love that.

What Fusion-io does is build an expensive, albeit way cool, memory board that can be addressed as a disk and stuffs it onto the PCIe bus of your physical computer. Its value-add is that it can stuff 100X the memory into the same physical footprint. It is a density play. I am totally cool with that.

Issue #1: It's memory. It's components. How can a little company, as great as it is doing, compete with bazillion dollar component manufacturers long term? Answer: it can't. Toshiba or Samsung or someone is going to ultimately do this and do it a ton cheaper than Fusion-io can. In 199 ish, I joined the hottest third party memory company in IT (to start a storage business) called Clearpoint. Clearpoint went from nothing to $54M (with zero VC money) on the back of MicroVax-II memory boards (16MB for a mere $9,000 or so). The company expanded into many other lines, shot up to $80M, I think. Wall Street came running to us. Two years later, the company was toast. Why? Because Clearpoint thought the competition was the OEM--Digital or IBM etc.--but it wasn't. The competition was from the people who built memory chips. They changed the game with SIMMS, which became commoditized so fast Clearpoint didn't know what hit it. $9000 boards at 95% gross profit sold for $900 and 5% gross profit 18 months later.

In the same timeframe, there was a little company in the MA area called ECC (or EEC--I can't remember). It built a memory subsystem that installed in the backplane that could be virtualized to become a resident SSD. Sound familiar?

Issue #2: In order for Fusion to win (in its current manifestation), it needs every physical machine to install its stuff. Once a customer does, in order to get proper utilization of that expensive asset (it cost as much as the server or more many times), the workload the customer runs is RESTRICTED to that physical server. Sure, the workload can be running on a virtual machine, but it HAS to run on that physical server. If you move that workload to another physical machine, you are, as my grandmother used to say, shit out of luck--unless that second machine also has a Fusion board installed.

Let's say that second server does have the Fusion board installed--the workload you cared about will still go wicked fast, so all is well, right? Wrong, because now your Fusion board in the first server is sitting there generating heat and sucking power just looking pretty, making the cost/utilization metric of the box the most atrocious in the data center. That isn't' going to win you any IT awards.

I'm 110% OK with the fact that we have certain application that absolutely, positively have to go a million kph (for my global pals) and have absolutely no issue with you solving that problem with a Fusion-io or equivalent type solution. Go nuts. What I am suggesting is that therein lies the issue for Fusion--it can never be mainstreamed into the world of IT unless A: it makes it so cheap that people don't care about putting one in every machine they ever buy and, quite frankly, it is simply not in a position to do that. B: if it does A, no one else decides to crush it and take that job for themselves. I find both unlikely.

The real issue in this is that in order for to be successful, Fusion-io is advocating solutions that are directly contrary to the natural forces of the market. It needs you to think DAS while the world is thinking networked storage. It needs you to think physical boxes while the world is disaggregating everything both inside and outside the physical world. That is just such a 1992 attitude.

What Will Happen:

Or perhaps, what should happen, and therefore will eventually happen? Simple: first, no one is going to leave Fusion-io alone. Do you know what NTAP's PAM is? It's Fusion-io (I don't think it's really from Fusion-io, but it acts and behaves the same way). This is not an area that big dogs are not paying attention to. Do you know what FAST is from EMC? It's IO director software that figures out in real-time how to place data between hyper fast flash memory and mega slow/cheap SATA disks--such that ALL of the IO to the box is totally and automatically optimized. Terabytess. Petabytes, not gigabytes. Do you think for a minute that EMC isn't considering extending that capability OUTSIDE its boxes? A networked cache above the backend perhaps? There is lots of activity going on. A server card (a.k.a., intelligent HBA) that might have some flash/cache capabilities on it? What if FAST ran throughout the whole data center, coordinating the placement of data in the optimum spot for EVERYTHING?

I'm getting ahead of myself here, but the point is this: to do wicked cool stuff with data, you have to have the ability to control, move, place, etc. everything from anywhere. You can't have ANY physical restrictions. So unless Fusion-io has plans to be able to leverage (via a network) their assets for ANY server from ANY server I can't see how this can be a life altering play. When I can utilize the ENTIRE collection of expensive, wicked fast memory that exists at the server (or storage, or network) no matter where is physically resides, then I really have something.

The Bigger Truth:

I really am not trying to shit on a company that has buzz times 100, has hit $100M in revenue real fast, and solves a real problem. I'm just trying to be a realist. It has the swagger and arrogance of a young Data Domain, but it doesn't have what Frank Slootman called the "long term secular trend" to ride here--it has a short term set of problems that by default WILL GO AWAY soon. Data Domain's trend was that data growth alone would continue to force more and more people to extend outside their backup windows. Nothing was going to stop that from happening. So he intersected that trend perfectly and solved the ECONOMIC problem of using disks (which solved the technical problem of the backup window) as a backup target by making dedupe a way of life. Billions in value later, the play still works. Data Domain was the rock star of the valley the same way Fusion is at this time, but it can't end the same way unless there is a whole lot that it is doing regarding what I just stated that I don't know about--which is in and of itself another issue. The people at Data Domain were cocky, but they were also smart: they talked to me/ESG and other smart people out there. They didn't think they knew it all, just the important stuff. They were right. I know firsthand that they sought and received value from those external strategy engagements. I know it made them money and accelerated success and I know that it helped them avoid mistakes they may have made. I also know they will be the first to admit it. (Do you see how I linked myself to their success? That was smooth.) Fusion doesn't talk to anyone as far as I can tell, which means they know it all. That's stupid. Pick your favorite best run company, and I guarantee you that one of the reasons it is as good as it is is that it KNOWS it doesn't know everything--and it does its best to comb the land for information and help. Companies that never look outside their own four walls at this stage are simply displaying ignorance, which is one of my most important success KPIs. I'm not smart, but I've got data.

Maybe it's personal, but I swear I don't think so!

Read more of Steve's blog entries at The Bigger Truth.