% fortune -ae paul murphy

A distant truth about "the cloud"

The near term truth about cloud computing is that it's only useful for enterprise applications if you're a user manager who wants to get around the obstacles in IT - that's what supported time sharing in the seventies and could make it a short term success again today.

Look beyond that, however, and the risks are simply too high for this ever to become a viable form of enterprise computing service. Google, a company with lots of resources and some pretty smart people, can neither keep gmail running consistently nor protect users from information abuse - and they're not even handicapped by any commitment to Microsoft's software.

Basically, my bottom line on cloud risk is that I'll put serious user data on the cloud right about the same day someone proves Dante's hell a physical reality, shows that it's frozen over, and proves that it's run entirely by squads of flying pigs.

On the other hand existing attempts at building a cloud do have two interesting uses:

  1. as a transit and filtering service for high volume data subject to resource intensive, but non lossy, initial processing. You might, for example, load a 10GB seismic log via a connection in Siberia, process it in the cloud, and download the 1GB filtered result with very little lag in Calgary; and,

  2. as an open source development and testing platform - providing quick access to wide variety of hardware and OS/combinations to project contributors around the world.

Beyond that, however, I think the standard cloud vision as some kind of universal computing resource is just eye candy for dilletantes - but that doesn't mean we won't see the evolution of a very different kind of cloud.

That one's in sight and coming fast - although, admittedly, today's google search stacks up to it just about the way Sony's 1979 walkman relates to today's ipod nano.

This is an eight minute video demo of something coming down the pike. The demo uses traditional input/output devices (real and virtual screens and keyboards), to communicate with the user and is, of course, canned in the sense that you can't actually get all the data they're showing in appropriate formats or at the rates they show.

But you should eventually be able to - and by then communication with that device should be based on something that looks a lot more like telepathy: at the very least, doing away with overt vocalization, keyboards, and screens.

When those devices become available, users will have the same problem the people making this demo did: where's the data coming from and who controls it?

This problem comes in two forms:

  1. on the survivability side, if you've ever tried to take away your teenage daughter's cell phone you'll be familiar with the kind of panic a device failure would cause among the information addicted. And that makes centralized storage an absolute requirement - meaning that you can use either the Sun Ray distributed server model or that model plus local storage, but you can't get away with just local storage.

  2. on the data source side the problems are sourcing, formatting and access: you'd pick up some data from local learning (for example, the name for your neighbor's new puppy), and some from traditional sources: google, twitter, linkedIn, and so on; but even if you had sources for everything, how do you format and access it?

The answer to both parts will, I think, have to be delivered via a cloud of data jobbers: people who provide personalized data support by running automated agents handling multi-point lookup, retrieval, evaluation, and reformatting for each of their customers.

Thus I might sign up for Gold Support with Bre-X data mining - meaning that my link would connect exclusively to Bre-X, automated agents at Bre-x would both collect and manage my data, and Bre-X partners in other jurisdictions would seamlessly take over agent services when I travel. As a result I could be completely senile and still pass a math exam in Moscow, cheerfully hand over my iLink to the nice policeman in Beijing, get a replacement at the airport in Seoul, and still expect to know what the customs agent charges when I land near Brussels.

Oddly we have a cloud like that now: it's called the telephone system - and really the bottom line on long term cloud computing is that data jobbing is the cloud opportunity - and it's their opportunity to miss.

Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.