The Road Not Taken – Robert Frost

TWO roads diverged in a yellow wood,

And sorry I could not travel both

And be one traveler, long I stood

And looked down one as far as I could

To where it bent in the undergrowth;
        

Then took the other, as just as fair,

And having perhaps the better claim

Because it was grassy and wanted wear;

Though as for that, the passing there

Had worn them really about the same,
 

And both that morning equally lay

In leaves no step had trodden black.

Oh, I marked the first for another day!

Yet knowing how way leads on to way

I doubted if I should ever come back.
 

I shall be telling this with a sigh

Somewhere ages and ages hence:

Two roads diverged in a wood, and I,

I took the one less traveled by,

And that has made all the difference.
 

     A well known poem, memorized by myself and my entire 3rd grade English class years ago.  The regularly missed point is that there really was no better choice, no better road, they were equal for different reasons, and only in blind hindsight can someone say one would have been better then the other. 

     Recently I was asked to talk about the reasons a storage start-up would chose to built a custom hardware platform vs. leverage a commodity server platform.  In reality, they both have merits, both, in early consideration, pretty equal. 

Servers as a Platform

     The commodity server platform as an appliance has always been an interesting one.  The benefits for a start-up are many.  Firstly, you get to leverage current technologies easily, toss in a new network card and you’ve got IB or 10GigE.  Secondly, open source Linux gives an instant development baseline, leverage the work others have done before you.  Lastly, your time can be spent on software development and features as a priority.  The complicated part is not getting the solution to work, but to get the most out of a platform not designed for IO, but for general computing.  The mix of software development between user space and kernel space, staying on top of security within a very open OS architecture, and overcoming the inherent single points of failure in today’s common server platforms, all become pain points very quickly.  However, for a cash poor start-up, the initial minimal costs and rapid development typically make up for choosing this road, especially if the exit strategy is change of ownership.  In recent years upstarts like Compellent (OEM’d SuperMicro Servers), LeftHand Networks(Dell, HP and a few other servers, although most recently ONLY HP Servers), Exanet (IBM, Dell, and a few other server vendors), DataCore (pretty much any server), and several others, chose the commodity server route and have had mixed success.

Hardware as a Platform

     The custom hardware road is much more complicated from the get go, but the effort up front can pay out big in the long run.  The upfront pain with custom hardware is running two engineering teams, one for the hardware side and another for the software.  The benefit of custom hardware is that chipsets, real-time operating systems, and form factors allow for HA hardware architectures, system built for performance and reliability, and ownership of every detail of the solution.  However, the effort to support hardware, roll with technology updates, and costs for manufacturing can burden even the most experienced engineering teams.  In mainstream storage the custom hardware route has been the standard of the likes of IBM, HP, EMC, HDS, and SUN STK, who cumulatively own the storage market. 

How to choose?

     In the end, how the technology was implemented is far less important then the impact of the architecture decisions to support the vendors choice.  The things I always pay attention to is the data path, what happens to the data path in normal state, degraded state, and a failed state.  The next area I will tend to look at is how has a company really leveraged either technology, was custom hardware required, or it is not a differentiator.  In the end, there isn’t always a clear winner, and doing a technology bake-off is always a good idea if the vendors will allow for it.