How can I top last week’s prediction about Google’s shipping container data centers? By explaining a bit more about how the system came to be and how it will work.
In last week’s column I told how Google has been experimenting with portable data centers built in standard 40-foot shipping containers. The idea isn’t new and it isn’t even Google’s. As far as I can tell it came originally from Brewster Kahle of the Internet Archive, who wants to replicate the archive here and there around the world and figured that a shipping container filled with servers and disk drives might be the easiest way to do so. Not only is it truly plug-and-play, but it is also a heck of a lot cheaper from a bit-schlepping perspective. Carrying a petabyte data center by ship from California to Australia is the virtual equivalent of an OC-192 optical connection – the world’s most powerful SneakerNet.
The Internet Archive sponsored presentations trying to drum up interest in this concept (the presentation, itself, courtesy of Bruce Baumgart, one of its authors, is one of this week’s links) and at one meeting Google’s Larry Page was in the audience. I guess he must have found something of interest.
At this point I should make a couple of confessions. First, I was off in my cost estimate last week by an order of magnitude. This error was pointed out to me by reader Paul Tuckfield and he’s right. That’ll teach me not to make quick-and-dirty calculations. I had the spec but no dollar numbers. But this actually makes the whole scenario MORE likely, not less, because Google likes projects that would be too great a financial risk for most others. It’s somewhere between $1.5-$3 billion, to implement, but who other than Microsoft or Yahoo could afford that? And who other than Google could pull it off?
Second, Matthew Glotzbach, senior product manager for Google’s Enterprise Division wrote to say that I made an error in the way I last week characterized Google’s Search Appliance. He said, “The product is a “plug and play” appliance as you said, and setup and maintenance requires little to no time or technical expertise. However, the setup and management is performed by the customer, and there is no communication between Google and the customer’s appliance. In the event of a problem or support issue, the customer can request that a Google technical specialist perform remote diagnostic and problem resolution, but at no time is the customer’s data pulled from their appliance back to Google’s data centers.”
This message from Mr. Glotzbach, by the way, was the only response I received from Google. So if you were expecting “We don’t need no stinking shipping container data centers” from Google, it didn’t happen. That’s important.
Getting back to Paul Tuckfield, he points out, too, that this whole idea of a network in a box (a very BIG box) is just the logical extrapolation of a trend. “I think the trend is there too, to package entire networks as a single system,” he said. “It makes sense because many applications do this already — websites and supercomputers being easy examples. The same trends of miniaturization that caused microprocessors to emerge from a background where supercomputers CPUs were built with discreet logic back in the 80’s will cause networks in a box to emerge in this decade.”
The big question is what Google will get for its $3 billion bet? Last week we covered the idea of doing massive video streaming or downloading through parallel peering arrangements. We also covered the basics of reducing latency for network-based AJAX applications to compete with Microsoft. But there is a LOT more.
Once you have a data center at every Internet peering point, you also have a data center in or near every major city in the developed world. That suggests Google might be interested in using the portable data centers for Voice-Over-IP telephony. Sitting 2-3 hops from every telephone and having available Google’s own fiber network and traffic shaping to give priority to its VoIP packets, Google could offer world-beating telephony performance, all for less than eBay is paying for Skype.
Another possible use for this parallel Internet is, if anything, more political than technical. The players in broadband Internet service – the telephone and cable companies for the most part – have as much to lose as they do to gain from the Internet. The telephone companies have at risk their voice service, which is already being undermined by VoIP. The cable TV companies are risking their video service as telcos get set to offer various DSL video channels. Each group realizes they can’t stop the progress of technology yet on some level each group would like to try. By grabbing a big fistful of optical fiber and having data centers at very peering point, though, Google offers an alternative in case one party or another is tempted to undermine the system through technical tricks like altering the packet interleaving to mess with VoIP as I have written before. Google’s success requires an open Internet and their presence and deep pockets guarantees that will be the case.
But the most important reason for Google to distribute its data centers in this way is to work most efficiently with a hardware device the company is thinking of providing to customers. This embedded device, for which I am afraid I have no name, is a small box covered with many types of ports – USB, RJ-45, RJ-11, analog and digital video, S-video, analog and optical sound, etc. Additional I/O that can’t be seen is WiFi and Bluetooth. This little box is Google’s interface to every computer, TV, and stereo system in your home, as well as linking to home automation and climate control. The cubes are networked together wirelessly in a mesh network, so only one need be attached to your broadband modem or router. Like VoIP adapters (it does that too, through the RJ-11 connector) the little cubes will come in the mail and when plugged in will just plain work.
Think about the businesses these little gizmos will enable. The trouble with VoIP in the home has been getting the service easily onto your home phone. Then get a box for each phone. The main hurdle of IP TV is getting it from your computer to your big screen TV. Just attach a box to every TV and it is done, with no PC even required. Sounds like Apple’s Video Express, eh? On top of entertainment and communication the cubes will support home alarm and automation systems – two businesses that are huge and also not generally on the radar screens of any Google competitors.
Throw a panic button atop every cube.
But for all this to work, especially with end-to-end elliptical encryption, you need a tight connection between the box client and a server, which is why those shipping containers need to be so broadly distributed and why Google will need so many of them, eventually numbering in the thousands to support hundreds of millions of cubes.
The cube, itself, is a sealed device literally embedded in epoxy. For a smart device, it is as dumb as Google can make it, because dumber is cheaper and dumber is less vulnerable to security breaches. For the cube, in addition to all its other functions, also handles Digital Rights Management.
Now imagine a world where Google Cubes were distributed as widely as AOL CD’s. It will be in Google’s interest to provide them in volume to every Google users, which is to say every broadband user everywhere. As a result, Google becomes overnight a major phone company, a major video entertainment provider, a major player in home automation and even medical telemetry.
The Google Box mesh network can reach out to nearby neighbors, too, bringing them onto the Internet in a way that would be difficult to stop or control even if the broadband ISPs wanted to, which they won’t, because Google will find a way to share the wealth with them.
It is not in Google’s interest to put out of business any ISPs, so they’ll try hard not to. But it IS in Google’s interest for there to be universal broadband coverage, which the Google Cubes will, for the most part, enable.
It reminds me a lot of Sun Microsystems’ vision for Jini, its Java-based distributed intelligence product that Microsoft derailed with Universal Plug-and-Play. If the Google Box is the practical equivalent of Jini, then it ought to be able to fulfill some of the grand plans for that technology, which was often presented as an intelligent sensor network to drive e-commerce. In the Jini model, a chip embedded in your car’s brake pads would signal that they were worn and needed to be replaced, but he network not only used that knowledge to turn on a dashboard warning light, it sent a message to the brake pad factory to build another pad, please. Well the Google Box is a coarser-grained version of that, with the tinier sensors being added later as technology comes down in price and the system pays for itself.
Is that a grand enough plan?