[daniel_levine]

This. I've studied traffic extensively and if you ever wonder how it works, it is actually quite similar to any other network. Imagine road traffic is the same as traffic through your servers. There are hard bottlenecks at various points, but in practice what matters is planning for variability. Real-life situations like someone cutting you off, people rubber-necking, the natural commute tendencies etc tend to be much more impactful. Unlike with servers, you can't just spin up a new lane, and "packet loss" is pretty much unacceptable :)

EDIT: For all you infra folks out there, imagine not being able to dynamically provision resources, no packet loss and worse, the packets can basically do what they want. That includes things like crashing into each other and stopping any other packets from making it through.

[jameshart]

re spinning up a new lane: the A38-M Aston Expressway into Birmingham, UK dynamically multiplexes its lanes together into variable width channels depending on the required upstream/downstream bandwidth :) - http://i3.birminghammail.co.uk/incoming/article1294911.ece/A...

[jbrooksuk]

I was watching Motorway Cops - or whatever it's called on BBC - just last night. There was a major crash, with a Mini and a Jaguar.

I didn't know that the direction of the Expressway changes, it's pretty cool, but has kind of put me off driving down it.

[colanderman]

There are highways in the US that do this with dividers. A specialized vehicle drives over them and shifts them one lane to the left/right twice each day.

[interurban]

A good example would be the Golden Gate Bridge (at least they used to do this, I haven't been to the west coast in several years).

[khuey]

Still does, but it's only over the bridge. The approaches have fixed bandwidth and can back up.

[CodeCube]

aww hell no. I'd be terrified of merging into that lane with no highway divider between me and oncoming traffic. Small two lane highways are bad enough! :P

[maxerickson]

There are less terrifying, less dynamic alternatives:

http://en.wikipedia.org/wiki/Barrier_transfer_machine

[CodeCube]

Nice! yeah, this would be preferable. If there's a catastrophe in the opposite lane, I'd really love it if it could stay over there while I'm going 70 miles an hour in the opposite direction :P

[andrewaylett]

Most of the time, there's an empty lane as a separator. When it is used, they don't allow motorbikes in it just in case, but it's really no worse than a two-lane undivided road.

It's also not necessary for you to use that lane if you don't want to -- it doesn't go anywhere special :).

[andrewaylett]

As a correction, as this road appears on the BBC's Motorway Cops regularly, and there was an incident featured in this weeks' episode (1): There is always an empty lane, which shifts depending on the time of day. Accidents are apparently very rare, and the speed limit is 50 mph, compared to 70 on a regular motorway and 60 on a 'normal' road.

[1]: http://www.bbc.co.uk/iplayer/episode/b03c7h9y

[aroch]

That is sort of terrifying but also quite cool

[dvanduzer]

So the older/well circulated article that this author complains about would be describing Van Jacobson's algorithm?

Am I missing something about what the author is saying about the empirical data?

The stark uniformity of the empirical data up to the congestion threshold compared to the wild variance in heavier traffic is exactly the point of the older article this author is critiquing. Of course congestion happens when you approach capacity. This article is rejecting the original's explanation of that variance, without offering any new explanation.