The Storage Revolution Everyone Should Join

Except Storage Vendors
It’s time to break the back of the storage oligopoly. Rip the still-beating heart out of this proprietary industry, drop it in a blender and hit “puree”. We have the weapons. History is on our side. Now is the time. Here is the plan.

The storage industry is badly broken. Vendors have gotten a free ride from the innovations of disk vendors to keep lowering array prices. Storage system interoperability is limited, storage networks are fragile, backups frequently don’t complete or are incomplete, margins are very high for what should be commodity products and storage management is hardly worthy of the name. Resellers and VARs, busy slathering on glue and bailing wire to hold things together, also get stuck with most of the evangelism, only to be relieved of their best customers when they are too successful.

Customers are paying a high cost as well. Not only in dollars, but in lost flexibility. CIOs are finding their jobs at risk as the Business Unit GMs look around and see smaller competitors rolling out high-value apps at a pace their own IT groups can only envy. Storage is the highest cost of most IT departments, who are paying $10, $15, even $20k a terabyte when disks are selling for $500 a terabyte. Vendors smother the big customers in support to keep them spending and to keep smaller, innovative vendors out.

With all this so profitable for vendors, whose hardware margins are the envy of the server world and whose thousands of point software products command high “value-based” prices, the industry has little incentive to change. And they don’t.

You Say You Want A Storage Revolution?
The good folks over at Storage Revolution have a plan: build open, agnostic storage management tools. A fine idea: many open-source projects are industry leading, like Linux and Apache. Yet, sadly, it won’t work here. Why? Let us count the ways:

  1. Historically, the glass house is extremely resistant to change. Name one revolutionary technology that took off in the glass house data center first.
  2. Conversely, new IT techologies are driven from users and BUs until the glass house is surrounded and forced to capitulate.
  3. Storage management is too easily sabotaged. Interface changes, new features, changing standards interpretations, there are 101 ways vendors can break “standards-based”. Don’t think they won’t.
  4. Commercial start-ups haven’t been able to do this even after spending tens of millions of dollars and hundreds of man-years of engineering. It is a lot of work. Aperi, anyone?
  5. Who will convince a datacenter manager to start using the product? How? They are highly risk-averse, for good reason.
  6. Finally, this attacks the wrong problem: managing a giant hairball. The management problem is a symptom, not a cause. The storage hairball is the problem.

We’d All Like To See the Plan
Like all good plans, this one is simple and uses the dynamics of the industry against itself while following the path of other successful insurrections. Industry can know exactly what everyone is doing while being unable to stop it. Expect lots of spitballs and the occasional (figurative, Mr NSA man) IED, as you start getting close. They’ll be a sign of success.

The work still needs to be done, but it will produce lasting change, just as Linux and Apache have. And the technical folks who lead the charge will have years of well-paid fun as the technology takes hold.

Proprietary Disk Arrays Are The Problem
The basic storage management problem isn’t everything, it is storage arrays. They all have their own intelligence, their own way of doing the things, their own APIs, their own wrinkles on every operational aspect. Pricewise, storage arrays are 90% armor, 5% cache and another 5% disk. The “added value” is totally out of control. Yet that “added value” is the basis for the entire industry’s ability to sell billions of dollars of management software. Rip the heart out of the array business and you’ve got a world where most of what we call storage management today simply goes away.

And that should be the ultimate goal. Storage is managed only because it needs to be managed. Build storage that doesn’t need management, or needs very little, and a new world opens up (no, not the one where the government records our actions 24 hours a day) where every person and business who could benefit by plentiful, reliable and inexpensive storage will have it. Just as broadband has benefited so many and created opportunities for millions.

So the plan is simple: build a highly-available, high-performance, low-cost, open-source storage array using a combination of open source and commodity products. Market it to Tell VARs and resellers about it, who are the natural channel to sell it to cost-sensitive SMBs. Let the rep build while more people swarm the technology.

Smart F1000 division GMs, tired of waiting for IT, will pick it up for their underfunded, time-critical, must-have apps. Then expect a struggling system or storage vendor will decide this is the next big thing that they can hang their hat on. Once that happens it is only a matter of time before everyone signs on. And, of course, put a standard management API on it.

Aren’t Storage Arrays Really Hard?
This is a bottom-up strategy. Start at the top like Zambeel, Cereva, Scale8 and, probably, Pillar, BlueArc and 3Par, and you are a fat target for wealthy, well-entrenched competitors with long-time relationships with the toughest customers in the industry. Differentiation is tough, and even with solid differentiation, breaking into F1000 datacenters is very difficult.

Don’t start at the top. Go for the SMB market, where your natural VAR allies have the distribution and the customers have the need. You’ll also get fast traction in technical markets, even in big companies, because IT is much more hands off with scientists and engineers.

Arrays are a lot less difficult to build than were even three years ago. Commodity dual and quad processor server boxes with PCI-Express I/O provide enough processing and fast connectivity to support many terabytes (Jonathan Schwartz, Sun’s new CEO, mentioned a 26TB configuration running on a four-way Opteron box). Toss in some clustering and you have a very sturdy hardware platform. Layer on the open-source 128-bit ZFS file system/volume manager/CDP that a very smart group of Sun engineers has spent 6 years working on, and you have an array that goes beyond 95% of what the industry is offering today —all for the cost of the hardware!

Frankly, when it comes to eviscerating multi-billion dollar companies with open-source products, it doesn’t come much easier than this. You can expect help from major chip companies, disk drive vendors, and enclosure makers, who’d like to see more of their product move out the door. Storage Revolutionaries: its time to storm the Bastille!

More Info
Learn more about ZFS here and here. I *think* Sun is using ZFS in Project Honeycomb as well.

Check out these ZFS benchmarks.

And here is a blogger’s transcript of Schwartz’s Q&A at ITXpo. The money quote:

So ZFS is a 128 bit file system–do you need 128 bits? Maybe not but you might need the 65th bit. Solaris with ZFS eliminates the need for RAID controllers…

So why did you spend several billion for a big iron storage company?