Sunday, March 24, 2013

Danny Hillis: The Internet Could Crash - We Need a Plan B.

Here's another interesting TED talk. The speaker is Danny Hillis, the designer of the now defunct "Connection Machine", a very parallel computer with a lot of lights on the front panels. One of those machines managed to show up as a backdrop in one of the Jurassic Park movies. When I was managing a computer center at Bell Labs, we did consider purchase of a "Connection Machine", but it seemed that my users weren't really ready for something way different from the Sun servers and DEC-VAX machines that were the heart of our operation at that time. A "Connection Machine" seemed too risky in any case since the price tag was large and the company attempting to sell us the system was not well established and therefore could (and indeed did)go out of business in the blink of an eye.

Danny Hillis: The Internet Could Crash - We Need a Plan B

The TED talk was given in February 2013.

Reminds me of the proposal I made to split up the computer center in Murray Hill Building 5. The assumption had been that in the event of a disaster that took out our computer center there, that we'd fall back onto the facilities of the "Murray Hill Computer Center" in Building 2. That might have been a feasible plan when our Building 5 operation was small relative to the size of the Building 2 computer center, but technology and user demand in Building 5 led to enormous growth in storage and compute capacity. It was clear to me that Building 2's computer center didn't have the oomph we'd need if something happened to Building 5. I found a small raised floor area in Building 4 that was slated to be decommissioned and proposed that we move some of our "eggs" out of the Building 5 basket. I wasn't promising that we'd be able to transparently absorb the loss of Building 5, just that it would assure that our users wouldn't be entirely out of business if something happened to one of the buildings.

I was especially nervous about Building 5 because it has a wood-truss roof protected by a water-sprinkler fire protection system. One time, when a pipe in the ceiling burst, Rick, our lead computer operator, described it as "Rick's VAX Wash" as torrents of water poured down on one of our VAX 11/780's. Much to my disappointment, I couldn't stir up any interest in our users who fund our budget in splitting the computer center partially out of Building 5. Ultimately I left that organization for a less political and more technical position elsewhere with Bell Labs, Murray Hill.

But God saw to "meting out justice". That winter heavy snows covered the roof of Building 5, and then rain saturated the accumulated snow, but temps dropped and turned the whole pile on the roof to heavy ice that refused to slide off the arched roof, a significant static load. The load was too much for the wood trusses and one of them snapped during the night, shifting the load to its neighbors, which in turn similarly snapped. In the morning, the only visible symptoms were that the suspended ceiling tiles inside the building were strangely out of place. This led to the building people figuring out what was wrong and evacuating the building. Desks were set up in the main lobby of Building 6 so people could continue to have a place to work. The computer operators were issued hard hats and Building 5 was closed to all but "essential personal". Rick called it "closed to all but dispensable personnel" until the trusses could be repaired several months later.

I admit to some small pleasure in quietly thinking, see, I told you we needed a Plan B. Alas, I suspect that Hillis's clarion call for a Plan B Internet is going to get as little support as my call for a Plan B for the Building 5 computer center. If it happens at all, maybe the Plan B Internet will be a side-effect of the transition to IPv6?? But if the IPv6 Internet is as vulnerable to attack as today's IPv4 Internet, that isn't much of a plan B as it merely requires a second attack to take down the 2nd Internet. Should Plan B look way different from the traditional international Internet and just focus on smaller communities (per country or per state, with paranoid gateways to get from region to region)? I don't see how to do it "safely" without having stronger "central control" than the Internet model has today. Big Government central control (ala the "Great Firewall of China") makes me plenty nervous, so I hope the network gurus of the world have some better ideas.

Please add your comments to this blog if you have suggestions on how to get a Plan B in place. What would it's scope be? What would it look like? How would you test it?