In Affinity Networking, Business and the Network, Enterprise Data Center, Featured, Industry Insights, Network Evolution, Plexxi Control, SDN

This week I had to travel to Dallas. Attached as I am to my favorite carrier, i took a connecting flight rather than an available non stop flight. And racked up a 3 hour delay as a result. I know I am not alone.

There are 499 commercial airports in the US. Out of these 499, about 40 of them are considered hubs, major connection airports for one of more airline. It is obvious that airports come in all sorts of sizes. From the ridiculously large like ORD, DFW, DEN, ATL and a few more, to the more modest hubs like BOS, PHX, CLT, to small regional ones. It is not by coincidence that airports are sized the way they are. While a portion is driven by physics (how much room do I have to build out an airport), a much larger factor is pure needed capacity, measured in the amount of flights that need to be served, the amount of passengers to be transported. And in places where we cannot create more capacity, multiple "smaller" airports combine to take on the load (JFK, LGA and EWR as an example). 

Together, these 499 airports serve a total of 16,509 towns and cities. On average, an airport serves about 33 towns and cities. The vast majority of travel is domestic, close to 750 million passengers are flown in and out of US airports every year. International travel is provided out of all hub airports and a few more of the smaller airports.

While we all complain about flying pretty much every time we fly, the overall system has over many years evolved into a mature and well structured logistics operation. Capacity in terms of airports, amount of flights, size of planes and routes between airports are constantly adjusted to the needs of the traveling public. When there is demand, new flights and routes are added. When planes are empty, the amount of flights is reduced, or routes are removed entirely. It's a game of economics, moving people costs money and the airlines and airports constantly adjust to maximize the opportunity to transport people and goods in a way that is as efficient as they can make it. It is a business, they worry about underutilized capacity, they worry about missing out on opportunities to move more passenger (and extracting money from us doing so). The combination of all air transport has a limit, but we find ways to maximize the amount of planes we can have in the air.

But like anything, there has to be a better solution. In the networking world we excel in creating connectivity, we have been doing it almost as long as the airline industry. I am sure that we can apply our many years of connecting things to air travel and get an even better solution. Air travel is quite similar to applications in a data center exchanging information. Airports are switches, moving traffic from one place to another. The towns and people served by them are the servers, the VMs, the flows that need to be moved, even the ratio of towns served by an airport is pretty similar to a typical rack of servers. As with airports and airlines, some routes are more traveled than others. Some airports have a lot of passengers traveling back and forth, certain related applications have lots of network traffic traveling back and forth. Hub airports are aggregation switches, moving lots of transit traffic, but also terminating travel as high volume destinations. They are also the airports where people and goods leave and return into the country, or network traffic leaving the DC.

With that in mind, I propose that we change the way we travel. The last few years in networking have clearly proven that there is a better way. First, I propose we create about 40-50 new super hub airports. Large transit airports, their only purpose to provide connecting flights, this airport is never a destination itself. We will place these airports in the middle of the country, about even distance from the rest of the airports. We then create routes from all the other airports to each of these hubs, and those are the only routes. We will use the same amount of planes, the same size planes, and the same amount of overall flights from all airports to each of these hub airports. We have a choice for the exact amount of capacity we create. We can pick the average need, or the maximum need. We will pick the average, fiscally conservative. Whenever you need to travel somewhere, you will have no reservation, or a specifically assigned flight. You go to your airport, and you get assigned to a next flight to any of the hub airports. When you arrive at the hub airport, you will be assigned to the next flight to your destination. There is lots of room at the transit airports for you to hang out, you may be in line a while before your flight leaves. And we apologize for the fact that this is probably quite a bit slower than a direct flight. Sorry. Oh, but if you travel in a group, you will travel on the same flights, thats the least we can do.

Without question this is extremely simple. Planning is simple. Creating the route network is simple. Operating it is simple. We treat everything the same. All airports are the same. Their capacity is the same. Cookie cutter. But its easy to see that this is painfully inefficient. Many airports will not have enough capacity to move all passengers and goods. People will be stuck at source or hub airports due to congestion, while other airports have empty planes leaving. Or we created maximum connectivity and tons of empty planes fly through the skies, airports are ghost towns, and money is wasted. 

Spine and leaf with ECMP may be simple, but just about every other logistics problem in the world is solved by delivering capacity where it is needed, when it is needed. It is solved by understanding patterns and building optimized topologies and paths based on it.  Capacity is not evenly spread. Not everything or everyone has the same travel needs. Logistics problems in the real world are Affinitized. Why do we believe a network does not have to be? We are using high school math to solve network connectivity. We are creating topologies and paths that our own brains can fairly easily reconstruct. The rest of the world uses far more sophisticated methods for calculating and creating optimized connectivity. We should too.

Leave a Comment

Contact Us

We're not around right now. But you can send us an email and we'll get back to you, asap.

Start typing and press Enter to search