Packets beat Circuits

So I got in yet another variation of the Mass Transit Discussion over the weekend. I have this discussion often enough that I decided to write up my position here, so that I can just point to it.

First off, you have to know a little history of telecommunications. (I’m a telecom geek, okay?) The big story of the last twenty years has been the total and complete triumph of various packet-switching architectures over circuit-switching.

Very simplistically, circuit switching started with Alexander Graham Bell. Your voice was converted to electrical signals in your telephone, then a pair of copper wires ran out to the street, where they were bundled with more pairs of copper wires, then finally to a telephone central office. At first manually, then automatically, and finally digitally, a connection was established between your pair of wires and the pair of wires terminating in someone else’s telephone, and you could talk to each other. State of the art for over a century, from 1876 through the 1980s.

Packet switching divides a signal (could be your voice — VOIP, could be Internet data, could be video, whatever) into multiple standard “packets” of information, then routes each packet independently towards its destination. A lot of the technology behind this was sponsored by the military, so that signals could “route around” damage in a wartime environment. Every bit of the Internet is fundamentally packet-oriented.

Does anyone remember ISDN? I didn’t think so. That was a circuit-switched architecture that sucked up billions of dollars in R&D, aimed at deploying a Basic Rate 64kb/s channel to every home and desktop. Lucent and Nortel (among others) finally got it working just in time to be steamrollered by the Internet. (For a fabulous look at how things fell apart ten years ago, check out Netheads vs. Bellheads at Me? I’m a Bellhead who learned better.)

Packet switching requires a lot more processing power than circuit switching… both at the edges of the network and in the core of the network. That’s why Alexander Graham Bell didn’t invent it, and that’s why the Arpanet/Internet was limited to 300 bps text for much of its early life.

But, courtesy of Moore’s Law, processing power is now free. And, with sufficient processing power, packet switching always beats circuit switching. For efficiency. For flexibility. For resiliency. For application diversity. For extendability. For future-proofing. Basically, anything circuits can do, packets can do better. That’s the lesson of the last twenty years of telecommunications, and it’s why Google is worth twice as much as AT&T. [Feb 2020 update: now 3.7x.]

Back to the Mass Transit Discussion:

Mass transit systems are circuits. Automobiles are packets.

Packet switching always beats circuit switching.

Mass transit (heavy rail, light rail, trolleys, busses) means capital-intensive routes with minimal flexibility. The train stops at a station, not in front of your house. Maybe you switch to a narrower-bandwidth circuit (in Atlanta, a MARTA bus) to get closer to your house. Then you switch to a yet-narrower-bandwidth circuit (your feet) to get all the way to your house. This is directly analogous to the digital transmission hierarchy where I spent ten years of my life… SONET circuits are multiplexed out of T3 circuits which are multiplexed out of T1 circuits which are multiplexed out of individual 64 kb/s voice channels.

With automobiles, on the other hand, every “packet” is dumped into the transportation network to be routed directly to its destination. This requires significant processing power. Luckily, most humans have more than sufficient processing power. (The ones applying makeup while talking on their cellphone while zooming down Highway 400 at 80 mph… well, I wonder about them. There are limits to multitasking for any CPU.)

So automobiles take you door to door. More importantly, they take you door-to-door-to-door-to-door… most people’s schedules aren’t just home-to-office and back. It’s home to drycleaner, to Starbucks, to office, to a lunch date, to the bank, to the office, to Little League, to home, to dance practice, to Applebee’s, to the mall, to home again. I defy anyone to navigate an itinerary like that in American suburbia using mass transit.

Packet switching always beats circuit switching.

“But they do it in New York City,” I hear you cry. Yes, because New York City grew up around mass transit. It’s physically different from Atlanta (or pretty much any other town in America outside the Northeast, except maybe Chicago and San Francisco). The circuits are dense enough to have connection points within walking distance, and the automobile infrastructure is expensive enough to discourage private cars. ($225,000 for a parking space? Bozhe moi!)

Atlanta grew up around cars. It’s fundamentally a packet-switched infrastructure.

Ask any telecom engineer. You cannot replace a packet-switched infrastructure with circuit switching for any reasonable amount of money. Can’t be done.

Look at the cities with successful public transit systems. With the partial exception of the Washington Metro (which violated the “any reasonable amount of money” proviso above), they grew up around rail systems; rail systems were not overlaid after the growth had occurred. (And I’ll note that even the Washington Metro does a lousy job of serving the high-growth suburbs to the west of the city.)

[Feb 2020 update: the Metrorail Silver Line finally opened in 2014 after 46 years of planning and construction. Note my 2007 comment about “any reasonable amount of money.]

So, as much as I love the plans for the BeltLine, and as much as I enjoy taking the Tube in London, spending public resources on mass transit in Atlanta is a waste of time and money.

But we have real transportation problems in Atlanta. Companies and entrepreneurs are reluctant to move here because of our legendary traffic. Millions of hours are wasted every year by Atlantans sitting in traffic. Cars are spewing megatons of pollution into the air we breathe. We’re the asthma capital of the USA. And all of the money we spend on petroleum is funding both sides of a war where people want to kill us.

Instead of trying to get people out of their cars, why don’t we try solving the real problems caused by cars?

  • Time-of-day pricing on toll roads — lots of toll roads! — would help even out the traffic jams. If you really have to be on Highway 400 between 7:30 and 8:30 am, it will cost you two bucks. Between 8:30 and 10:00, it’s a dollar. If you can wait until after 10:00 am, it’s fifty cents.
  • Add some lanes to freeways and major arterial routes. Charge for them. Convert HOV lanes to toll lanes. I don’t care if you call them “Lexus Lanes“… they’ll reduce congestion and reduce pollution. (HOV lanes make things worse.)
  • Higher gas taxes (and I’m talking an additional dollar or two per gallon) would encourage people to buy smaller more fuel-efficient cars. Not force them… if I really want to drive my 8 mpg sports car, I can do so, I’ll just have to pay more.
  • Put more timing lights on the entrance ramps. Build more roundabouts.
  • Getting serious about telecommuting a day a week would reduce traffic by 20% right off the bat.
  • Zoning changes would encourage more live/work/play complexes like Atlantic Station (hopefully with better security) and Glenwood Park.
  • Relaxing the rules against jitneys would allow more people to use the existing mass transit facilities. (Quick, ride MARTA to Cox Communications headquarters! Whoops, the Dunwoody MARTA station is a mile away. Don’t want to walk a mile in Atlanta heat? Sorry, jitneys are illegal here. Pull the car out of the driveway and add another vehicle to the I-285 traffic jam.)
  • Charge lots of money ($1000/year or so) for student parking at high schools. They can pay up, or carpool, or take the bus. (Yes, it’s a circuit-switch, but a circuit that’s intentionally routed in front of each student’s door. It’s horribly inefficient, but a high-schooler’s time isn’t worth much.)
  • FlexCar and ZipCar are great. If city planners really want to “Do Something,” subsidize more of those. [Feb 2020 update: FlexCar merged into ZipCar. Lyft and Uber didn’t exist when I wrote this post in 2007, but obviously, they would fit here.]
  • Subsidize recharging stations for plug-in hybrids.
  • Subsidize cellulosic ethanol refineries.

But for crying out loud, don’t spend billions of dollars on circuit-switched infrastructure that doesn’t address the needs of a packet-switched environment. You could probably do everything on my list above for the cost of a mile of MARTA heavy rail, and it would make a far greater difference in Atlanta’s quality of life.

Packet switching always beats circuit switching.

Those of us in Atlanta (and cities like Atlanta) are going to live in an automobile-dominated world for the rest of our lives. Deal with it, or move back to New York. Let’s make our city work better, rather than wishing for mass transit to do something that is fundamentally impossible.


  1. I guess this post just goes to show that a little knowledge is a dangerous thing.

    First of all, what’s happened with packet-switched networks – specifically, with TCP/IP and the Internet – is the development of a whole slew of middleboxes/middleblades, classification technologies, etc., to try and make the packet-switched network in essence back into a circuit-switched network. Firewalls, IDS/IPS, etc. are all attempts to do precisely that, as are many other mechanisms enterprises think they must have in order to gain some semblance of visibility into and control over their networks and the applications which run across them – extended to the best-effort Internet, in many cases.

    Secondly, I challenge you to come with me to Bangkok – a city which did not grow up around mass transit – and demonstrate to me precisely how automobiles are better than mass transit, since at rush hour I can take the Skytrain from the Sukhumvit station to the Chit Lom station during rush hour and arrive within 15 minutes of my departure, vs. 30 minute plus in an automobile.

    Both the premise and the analogy are flawed.

  2. Anonymous says

    … except that there is no analog of ‘virtually free virtually unlimited computational power’ in the transit grid, which is the reason that both people and goods are more efficiently transferred on buses and trucks rather than in cars. Imagine if UPS put one package in each vehicle! Could they possibly break-even in that scenario? Certainly not.

  3. Anonymous says

    How is mass transit like circuit switching? If passengers individually decide where to get on and off, isn’t that exactly like packet switching?

    I’d think if you were to draw an analogy between a mass-transit system and a telecom network, the mass-transit system would be a packet-switched connection with enormous bandwidth but poor latency.

    Don’t packet-switched broadband connections have exactly the same bandwidth heirarchy that you ascribe to circuit-switched networks?

  4. halojones-fan says

    anonymous 2:
    “How is mass transit like circuit switching? If passengers individually decide where to get on and off, isn’t that exactly like packet switching?”

    Yes–if there are only four websites in the entire world.

    “Secondly, I challenge you to come with me to Bangkok – a city which did not grow up around mass transit – and demonstrate to me precisely how automobiles are better than mass transit…”

    It’s an analogy, not an exact description. You can have an infinite number of packets going through the same stretch of fiber/cable/air/whatever.

    And that particular failure doesn’t take away from his point–which is that the most efficient system is one where you get in your car as soon as you decide to leave and drive directly to your destination. Anything beyond that adds time and effort with no productive result. Ergo, the appropriate response is to increase network capacity to allow more “packets” to be present at one location at one time.

    I will say that academic vc doesn’t quite follow through–his proposed mitigations for Atlanta’s traffic are same-ol’-same-ol’. His chief assertion, though, is that it’s inherently less efficient to spend money on mass-transit than to spend it on automobile transit; and I certainly agree.

    One thing I’ll point out is that the only place I’ve seen an effective mass transit system, it had so many individual trams/cabs/buses/whatever that it was about as available as a personal car; and it was in a very small region, so small that there were “only four websites”.

  5. halojones-fan says

    PS — the failure of the “car-pool lane” concept is easily seen here in California. The state legislature decided that it would be all enviro-sexy if they allowed Prius owners to purchase stickers allowing them to drive in the car-pool lane. They recently ended the program when they learned, to their embarrassment, that the car-pool lanes had as many single-driver Priuses as they had multi-passenger cars…

  6. Great post Stephen. This is a terrific comparison. I would add one more item to your mix – strong support for telecommuting.

    Something I have always wondered about for all of those who think that NYC has the world’s perfect mass transit system – why is the island practically sinking under the weight of all of those yellow taxis? Hmmm.

  7. Stephen Fleming says

    at rush hour I can take the Skytrain from the Sukhumvit station to the Chit Lom station during rush hour and arrive within 15 minutes of my departure

    Well, if you want to go from a place that’s walking distance to Sukhumvit to a place that is walking distance from Chit Lom, well and good. I suspect there are quite a few destinations in Bangkok in which your claim would be laughable.

    (Same is true in Atlanta… if I want to go from BellSouth’s Midtown office to the airport, I’d be a fool for not using MARTA. From my office a few blocks away, it’s a tossup. From most offices in the suburbs, using MARTA adds significant hassle and delay compared to just driving to the airport.)

  8. Stephen Fleming says

    both people and goods are more efficiently transferred on buses and trucks rather than in cars. Imagine if UPS put one package in each vehicle!

    Your claim of “more efficient” is only valid if the value of time for the package/person being transported is zero.

    Note that for packages where time is important, there are courier services that do indeed carry one box per vehicle.

    Most boxes don’t care if they’re locked in little windowless trucks for most of a day. People do. At least, I do. I value my time highly, which is why I value my personal transportation flexibility.

  9. Stephen Fleming says

    I will say that academic vc doesn’t quite follow through–his proposed mitigations for Atlanta’s traffic are same-ol’-same-ol’.

    I’m interested in your suggestions. Some policymakers have privately emailed me asking for more discussion, so this may not be an entirely theoretical exercise. (Since it isn’t obvious to some visitors to this blog, ‘academicvc’ and ‘stephen fleming’ are one and the same… that’s me!)

  10. I like the comparison, but there is a limit to the argument that needs to be added, and that’s bandwidth. The number of cars/hour that can crawl down a roadway is the bandwidth limit. The same road has higher bandwidth for buses than for cars. Rail bandwidth is INDEPENDENT of the load up to a very high level, and the opposite is true for roadways.
    So, unitl the “packet switched” system saturates the bandwith, it is inarguably better. At some point, packet “collisions” (heh) dominate and the packet switched system is much slower than circuit switched. In the real world, a blend is best, because the cross over points in efficiency are complex, and location and time of day dependent.

  11. Anonymous says

    One aspect of packet switching that you have not attempted to carry over to transportation is the way we deal with congestion. For those who are not aware, when the queues in a router get too long, packets are randomly selected for deletion and dropped without notification. The senders and receivers are supposed to realise that the failure to receive a packet signals congestion and react accordingly. The details of this depend on the actual protocol in use.

    So the equivalent in car terms would be that when the queue at the traffic lights gets too long vehicles are randomly selected, dragged to the side of the road and put through a crusher. I would suggest that if this was implemented, congestion would cease to be a problem quite quickly, particularly since logicaly the crushing would include all the contents of the package. there might however be other problems and I am sure there would be many objections.

    OK, leaving aside the unworkable aspects of that, if you can’t drop packets, you must prevent them being sent. By maintaining the traffic in the most congested part of the network just below the point of collapse you maximise the traffic through it. If the traffic gets too high you get gridlock, or the electronic equivalent. So a more workable concept might be that you notify the system of what you want to do and it then tells you when to leave and which route to follow. Perhaps to allow for contingencies and emergencies such a system might allow you to bid for a slot at a higher price, in much the same way that an airline ticket purchased in advance is likely to be cheaper than one bought just before departure. Implementation of this would not be easy, since not only would it have to allocate capacity for planned trips, it would also have to reallocate on the fly to work around problems in the network, such as accidents, traffic light failures and so on.

    I also doubt that many would be happy with a system that would at times effectively tell you to forget it when you put in a request to make a trip. Although it could be argued that the congestion at times is effectively telling you the same thing.

  12. BlacquesJacquesShellacques says

    The post is quite correct. The naysayers are wrong and have the usual tendency towards socialist totalitarianism.

    -Buses or cars
    -Mainframes or PCs
    -‘5 year plans’ or the free market.

    The most powerful of the mathematics is statistics. Einstein was wrong, God does play dice with the universe and
    laughs the whole time at ponderous central planners.

    “Ride ze fucking subway or ve vill send you to ze Gulag” is inherently funny.

  13. Anonymous says

    Your central analogy is clearly entirely backwards.

    Like circuit-switching, car travel takes a passenger from point-to-point directly. If the infrastructure is massive enough, it’s the highest-performance option. Like circuit-switching, it’s also inherently inefficient and expensive.

    Like packet-switching, mass transit takes a passenger through a number of pre-planned, interconnected routes, mostly stopping in places where the passenger doesn’t actually want to be. Although it’s lower-performance, more complex and less reliable than point-to-point transport, it makes more efficient use of whatever infrastructure is available.

    That’s why packet switching is popular and circuit-switching is unpopular. Not because packet-switching is inherently higher-performance (it’s not) but because it offers better bang for the buck.

    Transportation and telecommunications are probably different enough that this sort of analogy is of little use either way, though. The transport problems we face are hardly abstract ones. Why bother with analogies?

  14. I’d like to acquaint you with a technology that is close to your packets analogy. It is called Personal Rapid Transit. It uses automated, electric vehicles that provide station-to-station non-stop travel on a station-rich urban network. It’s an idea that has been around for some time but now appears to be gaining considerable interest around the world.

    The first PRT project is now underway at Heathrow Airport. A second project has built a test track in Uppsala, Sweden and will soon undertake a rigorous testing program with multiple vehicles. The Masdar Initiative eco-city project in Abu Dhabi plans to provide a PRT circulator system for its auto-free, zero-carbon, new city.

    I’ve collected a lot of information about PRT and its available at:

    Door-to-door at both ends is certainly desirable. One such innovative system is being developed in Denmark. It is called RUF and is know as a dualmode system (vehicles can be operated on an automated guideway as well as on conventional streets). is their website. Other dualmode systems are under development and can be viewed at my dualmode webpage:

    Perhaps one or more of these technologies could help the Atlanta region work its way out of the transportation mess that I read about from time to time. I’m looking forward to your 2nd article and hope that you can find some time to examine and assess PRT and dualmode as possible transportation solutions that are badly needed in our cities and others around the world. MASS transit is not the answer for the dispersed metropolis. Millions of additional personal autos, no matter how green, aren’t either.

  15. The URLs in my previous post are:

  16. fruminator says

    There’s very little I do know about in the world, but computer science and mass transit happen to be two things I like to consider myself at least somewhat knowledgeable about. There may be lots of reasons why cars are better than transit in lots of situations, but the analogy to packet and circuit doesn’t convey any of them.


  1. […] What the Heck Do I Want? February 27, 2012 Leave a Comment Derek Edwards found my ancient post Packets beat Circuits and wrote this response. I replied in his comment stream, but decided to duplicate my comment here […]