Friday, March 09, 2007

Why Packet Networks are More Reliable than the PSTN

Hard to believe? Give me a minute of your time.

In a recent post, Patrick made a comment about Van Jacobson's work at Cisco, and provided a link to a recent presentation called "A New Way to Look at Networking." I looked at this video a while ago, and it was a watershed moment for me. If you are unfamiliar with Van Jacobson, here's a section from Wikipedia:
Jacobson is best known for his work in IP network performance and scaling; his work redesigning TCP/IP's flow control algorithms to better handle congestion is said to have saved the Internet from collapsing due to traffic in 1988-1989. He is also well-known for the TCP/IP Header Compression protocol described in RFC1144, mainly meant to improve performance over low-speed links, popularly known as Van Jacobson TCP/IP Header Compression. Furthermore he has co-written a few widely used network diagnostics tools, such as traceroute, pathchar and tcpdump.
Now, Van Jacobson is working on the next generation Internet. In particular, the current generation of IP networking is based on the ideas of peer to peer communication, where pair wise conversations dominate the usage patterns, and we focus on the endpoints. The next generation of internet needs to focus on the data, not the endpoints. I deeply recommend that, if you are CTO or technical professional in the communications field, that you watch this video and commit it to memory.

In this talk, he makes an excellent point, which I want to amplify here. Since the topology of the phone network depends on concentrating ten thousand pairs of wires at the central office, it became really important to make the equipment at the central office bulletproof. Why? Because you had a physical concentration of wires at one place, and it was basically impractical to run a second set of wires somewhere else. Notice that in packet networks, this isn't the case, and there are several paths to the same point. The PSTN was basically unreliable, so you needed to make the elements reliable. In data networks, this is exactly opposite.

The business ramification? You have Verizon's thousands of central offices full of multi-million dollar switches. Yahoo! Thousands of $500.00 PCs. When a Verizon switch goes down, you hear about it on the 6 O'clock news. When a Yahoo! PC goes down, well... who cares? Lots of other servers to serve you that web page.

No comments: