Reading that, years ago, was what led me to figure out how to approach congestion control. Everybody was reading Kleinrock and assuming random arrival rates. Klienrock had done a study of Western Union Plan 55-A, which was Sendmail for telegrams, built with paper tape for buffering and telephone switches for routing. That led to the original analyses of how the Arpanet should work.
The trouble with that approach is that it assumes packets just arrive randomly no matter what the network is doing. That was true for Plan 55-A; delays in the network had little influence on the submission rate for new telegrams. But it's not true once you put a protocol with backpressure, like TCP, on top of the raw packets. Now congestion becomes a feedback control system. I figured that out in the early 1980s and wrote RFC 970.[1]
The idea of having one big line was introduced by banks some time in the 1970s. It helped some. But the big breakthrough came from Walter Wriston [2], CEO of what is now Citibank, who pushed his people to develop in-bank terminals and then ATMs, to get rid of the lines entirely.
A classic piece of advice from retail consultants is "never place an obstacle in front of a customer who is ready to buy". Amazon's "one-click ordering" is the classic on-line example. Gap stores used to be noted for getting this. They had big, empty counters and more than enough people ready to check customers out. All that crap retailers put at checkouts? Few customers buy it, and it limits how much merchandise the customer can put on the counter.
The big queuing theory insight for line management is that if you don't have some idle time, in an open loop situation the line length will go to infinity. In the real world, you start to lose customers. That's how closed loop feedback reacts to retailer incompetence.
"All that crap retailers put at checkouts? Few customers buy it, and it limits how much merchandise the customer can put on the counter."
People select less stuff when they're shopping because they're concerned it won't fit on the counter, or they take a bunch of stuff there but end up not paying for it because it won't fit, or what?
The trouble with that approach is that it assumes packets just arrive randomly no matter what the network is doing. That was true for Plan 55-A; delays in the network had little influence on the submission rate for new telegrams. But it's not true once you put a protocol with backpressure, like TCP, on top of the raw packets. Now congestion becomes a feedback control system. I figured that out in the early 1980s and wrote RFC 970.[1]
The idea of having one big line was introduced by banks some time in the 1970s. It helped some. But the big breakthrough came from Walter Wriston [2], CEO of what is now Citibank, who pushed his people to develop in-bank terminals and then ATMs, to get rid of the lines entirely.
A classic piece of advice from retail consultants is "never place an obstacle in front of a customer who is ready to buy". Amazon's "one-click ordering" is the classic on-line example. Gap stores used to be noted for getting this. They had big, empty counters and more than enough people ready to check customers out. All that crap retailers put at checkouts? Few customers buy it, and it limits how much merchandise the customer can put on the counter.
The big queuing theory insight for line management is that if you don't have some idle time, in an open loop situation the line length will go to infinity. In the real world, you start to lose customers. That's how closed loop feedback reacts to retailer incompetence.
[1] https://tools.ietf.org/html/rfc970
[2] https://en.wikipedia.org/wiki/Walter_Wriston