Ken Calvert
This was a very interesting presentation from Ken Calvert of the University of Kentucky. Rough notes:
He's a systems guy, an engineer. His first question in response to “what would you do with a gigabit per second” is “what do you mean?” Would this be shared v. dedicated? wired v. wireless? local area v. last-hop vs. internet wide? will there be another bottleneck elsewhere? can I do whatever I want, or are there acceptable use constraints (no servers)? is it symmetric?
He focuses on technology limitations. What limitations exist inside and outside the home that might be bottlenecks on 1gig? The serial ATA interface is okay – 1-3 gigabit. The latest digital display is fast enough. Memory bandwidth is high enough too - 10 gig/second. The home lan is 1 gigabit ethernet. So we can do this inside the home.
The big problem is the last mile. Right now, we only have about 1-10% of what we need to have a gig a second. In the core of the network, there are no bottlenecks. Same for wireless — WiMax and WiBro are also 1-10% of 1 gig/second.
So let's assume we have 1g last mile/symmetric, and a gig in the home. Memory bandwidth may be a constraint for high-throughput applications requiring significant software processing. They need 3.5 times bandwidth because of redundancy. But that's not really a big issue because we have big caches in our systems — big enough to hold all the data.
Protocol bottlenecks? Not really. Right now TCP is 70%, Bittorrent the rest. TCP's congestion control is wasteful of bandwidth on startup by sending very few packets per roundtrip. So we could trade bandwidth for latency. He suggests we could, instead of waiting a round trip time to learn which of several possible events occurred, send all possible reactions plus policy for choosing. So — send all the options, get response back for one that actually happened. Trade bandwidth for latency.
There may also be local distribution bottlenecks — most homes aren't wired for gigabit speeds. Gateway/router — that'll show up. But people will need upgrades. Content — interactive immersive environments, 3D conferencing.
But he sees an interesting trend in display interfaces. VESA is the org that standardizes display into PC — DVI. DVI is sophisticated, assumes that display has a fair amount of intelligence. But the interface is continuing to happen at a higher level
Eg, DisplayPort allows us to hook PCs up to same display. Now there's net2display — has IP interface to display. You don't need a PC. It's just a smart display. The services/applications can be anywhere in the network, and the apps could be far away. So you walk up to it with your computing device, and the human-size display will do the work for you.
Visionary part: assume a ubiquitously available, local area gigabit wireless channel (bluetooth 1gbps). The vision: separate computing devices from input/output devices via an open, networked interface. This will allow the separation of computing from input-output devices. This frees up PC and portable computer from having to have an integrated display. He doesn't think you can do real work through a phone interface, but a phone-size package will hold a good machine if you take away display and keyboard! Display is separable. You'll have protocols for dynamically discovering and establishing secure associations with such networked human interface devices. Then each can evolve independently. He calls this “belt-top computing.”
You'll have a permanent personal mobile computer, internal storage, single input/output channel, wireless interfaces, peripherals connected via network, human sized interface devices.
Comments
Got something to say?