The capacity illusion

Recently, the always thoughtful CTO of Hitachi Data Systems, Hu Yoshida, had a couple of posts (here and here) that point up a dilemma. In one, a CIO is decrying the 20% utilization of his data center storage, and in the other Hu is marveling at the fact that storage is so cheap internet entrepreneurs can give it away for free.

So which is it: cheap enough to give away for free or costly enough to worry about low utilization? The answer, of course, is both. Data center storage is expensive, and the internet storage is cheap.

Yet the interesting part is that both focused on capacity, the cheapest part of storage by far. How cheap? Over the last couple of decades disk storage capacity has grown by 100x per decade while price per byte plunged. Price per IO has decreased about 10x. IO is steadily becoming much more expensive while capacity is steadily becoming much cheaper.

The Money Illusion
Economists talk about the Money Illusion, where people feel like they are better off because they are earning more dollars, even when those dollars buy far less.

The Capacity Illusion is that capacity is somehow the most important metric that storage systems, and their managers, should be measured on. Baloney! The expensive component in storage is the I/O.

Run the numbers
Five years ago, the average disk drive cost about $4/GB while the average cost of OLTP tpmc was about $20. Today, 3.5" disks are about $0.30/GB and OLTP tpmc is about $4. So capacity is less than 1/10th the cost, while I/O is about 1/5th the cost. The relative cost of I/O has doubled in the last five years, even with all the fancy optimizations DBMS vendors and system architects have implemented to conserve I/O.

Of course you don't want your Exchange server running out of capacity. But capacity needs don't lurch from high to low and back again. It is I/O's bursty nature that makes configuring for I/O first the smart thing to do.

On the last day of the quarter you don't want your order processing choking for lack of throughput. Since IOPS vary so much, and since periods of high IOPS often correspond to periods of high business activity, it is much smarter from both a business and an IT (i.e. CIO job security) perspective to focus on IOPS instead of capacity.

Waste what is cheap to preserve what is dear
It is high time for the vendor community to get its collective head out of the sand and start pushing IOPS, not capacity, as the appropriate metric of goodness in storage. I got a lot of comments on this over at StorageMojo.com and the saddest was "so how do we know what our I/O requirements are?" A fair point and an indictment of lazy and complacent vendors that don't even provide the tools customers need to understand their environments. You might also have your engineers look at using cheap GB to conserve expensive I/O.

But safety doesn't sell!
That was Detroit's mantra for years as they resisted seat belts and air bags. I can just hear the folks in Hopkinton and Palo Alto snorting derision as well. Folks, if you don't focus customers on what is important instead tail fins and two-tone paint jobs, they will wake up one day and realize they've been had. And there goes your expensive branding and customer loyalty. Just ask Detroit.