the Onda

This year at the D conference Michael Dell was asked whether he was worried that the shift towards cloud computing would affect his PC business. Despite not usually seeming like a visionary, Dell gave a great answer. He said that over the last 20 years, every time bandwidth increases, it was his observation that so did the need for processing power on both sides of the pipe, and that because of that, he felt pretty good about the future of his PC business.

Just look at the iPhone or the N95, two mobile phones that pack a tremendous amount of processing power per ounce of weight. Despite being even more suited a a class to offload work to servers in the cloud, smart phones as a category seem to be growing more powerful in their display technology (hardware accelerated video), processing technology (Intel Atom), and general peripherals (5MP cameras, GPSes). Not only are these devices being packed to the gills with more transistors than a mid 1990s PC, but developers are rushing to PC-like development environments like the iPhone's and Google's Android to take advantage of the additional horsepower instead of just writing web applications for the increasingly more powerful web browsers that come with these things.

And it is not just about local processing power; the latest issue of Wired has what will no doubt become a classic piece by Kevin Kelly on the emergent distributed 12-million-teraflop computer that all of our gizmos are getting wired up to make. In the piece, there is a great chart that quantifies the shipped quantities of various different devices with CPUs at their core: from PCs to DVRs, from cellphones to cameras.

Now everyone knows that there are roughly 3 times more cellphones than PCs in the world today, but the stat that I found more interesting is that there are 44 times more PCs out in the wild than servers. Though I realize that it is probably difficult to define what a "server" is in today's world of quad-core x86 machines, the magnitude of that difference brought to mind the delta between storage at the client tier (in offices, in people's homes, at school), and the storage "in the cloud" (i.e., S3).

Even if you assume that the typical x86 server has 13x more storage than the typical PC (a terabyte of addressable storage versus a measly 80GB because you have to factor in the installed base more for the PCs), you are still talking about something on the order of about 100 million petabytes for the client tier and less than a third of that for the cloud tier.

Anyone who has tried to back up a photo collection to a cloud service like .Mac, to say nothing of a music or video collections, knows this at a gut level. The challenges with storage in the client tier have always been consistent addressability and reliability, but in a replicated and distributed world (a la Kelly's megacomputer), we might just be able to make better use of all of those petabytes.

Processing and storage made the PC revolution the juggernaut that it has been. It is why we've come to expect the interactivity native application developers running into the smart phone space are clamoring for, and it's why the dark matter of today's computing environment is composed of billions of hard drives, powered and accessed in a massively distributed way.

It is going to be a while before the cloud catches up with that (datacenter economics and bandwidth being what they are), and until it does, we might all be careful of falling elephants.

Post a comment