the onda

When two people as different as Larry Ellison and Richard Stallman manage to crap on a trend in the same week, it makes one think. In this case, the excrement was directed at the notion of "cloud computing," or moving what has traditionally happened on native PC applications to a server in a data center and a browser-based client. As Ellison stated, it's become such the rage that it's almost more helpful to describe the things that tech companies are doing these days that are *not* classified as cloud computing.

In thinking about the return to centralizing workloads, I was wondering whether we're just on that technology yo-yo that drives everyone from one extreme to another— from totally distributed to totally centralized and back. After all, computing did start in a completely centralized way and most of its history has been towards distributing the workload. Here is my napkin art on the trend:

The evolution of computing

Now I'd be ready to sign up to the theory that we are going to yo-yo here as well, riding that red curve back and forth till the end of time— but for the blue line of ubiquitous bandwidth, a trend which seems as unidirectional as Moore's law.

In each of the previous shifts in computing (from mainframes to minis to PCs), it would seem that the driving force was an democratization of computing resources, with equivalent capabilities being offered to an ever-expanding audience. The businesses who couldn't afford mainframes thought the world of minis, and the consumers who couldn't afford those were delighted with the limited PCs of the day. In the case of browser-based computing, the democratization is taken one step further: think of the developing world where an MS Office license is ridiculous but where free Google Apps at an Internet cafe is a complete bargain.

This trend makes the notion of the yo-yo rolling back the curve pretty unlikely, unless of course one of the following factors makes distributed computing compelling again:

Features: There are still apps that can not be shoved into the browser: video editing or hardcore gaming. The advent of richer browsers (think Chrome with its 10x faster Javascript implementation) is definitely fighting this trend, especially for all but the most specialized of computing tasks (video, music, CAD/CAM, programming).

Cost: It's hard to imagine any scenario where the economies of scale don't favor huge data centers, especially when we add the cost to administer computers, but it might happen. Imagine solar-powered computers for instance— since rooftop surface area is much less plentiful in a datacenter, there might be some advantages where the lack of density actually helps the overall economics. Heat is another factor as is anything else that is currently a limiter inside a data center.

Policy: Think of the privacy implications of Google storing all of your data, or Facebook having access to your entire addressbook. It seems unlikely that we won't get much more prickly about privacy going forward, and it is not inconceivable that we won't see policy-driven decisions emerge that might force some degree of de-centralization. For instance, I can't imagine government data being floated into a cloud— and especially not a foreign government's data into an American company's data center.

I'm not sure there is anything in these three buckets that will ever get us back to the ultra-distributed world that the PC ushered in in the late 1970s, so I just hope that we're not a frog in a pot that is slowly having its temperature raised— not even aware of the boiling we are about to experience.