|By Marten Terpstra||
|December 6, 2013 03:15 PM EST||
I work mostly in Plexxi’s office in Nashua, NH. That is about a 12-minute commute for me, backroads only through two sleepy towns. Very convenient and a great improvement over my previous commute. Every now and again I make the trek out to our Cambridge office and it is painful. About 40 miles and around 45 minutes of mostly highway on a good day; early morning this is easily a 90-minute exercise, and finding myself on the road for two hours is not unusual.
I am not very attached to a GPS, but whenever I travel a distance or know I am going to hit traffic, I turn on Waze on my phone and let it guide me to where I need to go. Waze has only failed me twice in getting me where I wanted to go (for some odd reason it brings you to a corporate airport when asking to go to DFW), but I am usually quite impressed with how it gets me there. It will almost always adjust the route along the way based on traffic, and it is rare that I use the exact same route to get to the Cambridge office.
Stuck in traffic earlier this week for a different reason my mind started wandering. If all of us used Waze (not pushing Waze specifically here, it just happens to be my GPS app of choice), would the combined commute time for all of us drop? Waze uses a variety of feedback mechanisms including current speed and user-reported traffic jams, accidents, etc. If we would all blindly follow what this application would tell us, would its global view of the state of traffic and where people needed to go lead to an overall better commute experience?
Interestingly (and probably not surprisingly) there is quite a bit of research on feedback-based traffic management. Traffic management in its non-network sense that is. There is this paper that looks at speed control based on traffic feedback and applies serious math (that seems to be a theme here at Plexxi) to try and prove whether it can be done safely. This paper discusses the use of GPS to collect road utilization, speed and other variables that can be used for route construction. Google some more and you find lots of variations of GPS-based traffic monitoring proposals, patents and academic papers.
There is a bit of a common theme in all of these discussions. Having active feedback and a centralized view and orchestration of what needs to go where will provide an overall better solution. All traffic on the road interacts with each other, and centrally orchestrated routes, speeds, traffic lights, etc., will provide a better overall solution compared to individual decision making based on only on your immediate surroundings. And we certainly have the compute power these days to work through the math to get results.
It is clear that we believe in those overall thoughts for network engineering too. Plexxi Control creates a L1, L2 and L3 network infrastructure based on where traffic is coming from and where it needs to go. And how much of it. And whether it needs special treatment. When calculated, it informs the drivers (the switches) how to route traffic, and will receive utilization feedback so it can make even better choices next time.
This does not mean that we believe in the centralization of all things networking. Switches are intelligent devices that can and need to make their own choices based on their immediate environment. When a link does down (like a accident on a highway), the switch needs to have pre-calculated ways to re-route the traffic. But all these choices are made within the global context provided by the controller.
I am fascinated by the various videos and demonstrations of quadcopters that cooperate to do certain tasks. This one is specifically impressive. Even in those systems, there is centralized control that tells the quadcopters what to do as an overall goal or task, but then use local intelligence and environment feedback for specific movements and adjustments. That is exactly the type of model we believe in at Plexxi. We want the controller to specify the overall movement of traffic but allow the switches enough intelligence to place to deal with failures in the routes calculated, even which specific traffic to place on which segments of multi-path routes, and perhaps in the future provide it with enough policy information to change forwarding patterns based on momentary utilization.
Yes, I think the chances are pretty good that if we would all use Waze or equivalent GPS and followed its directions blindly that we would improve overall commute times. I am sure it would need some tuning, but as an overall premise, I am a believer.
[today's fun fact: a GPS satellite signal takes between 65 and 85 milliseconds to reach earth. <insert can't-resist-comment-about-the-latency-of-multi-tier-network-architectures here>]
To read more on this topic, check out: