John Day: IP Networks (v4 and v6) are Fundamentally Flawed
I was at FutureNet Expo in Boston last week where I saw a very jarring and interesting presentation made by John Day, a professor at Boston University. If you’d like to take a look at the presentation, you can find it here. John has been involved with some fairly significant projects that were responsible for the Internet we have today. He is also founding member of the Pouzin Society, an organization named for Louis Pouzin, the father of the datagram and designer of the first packet communications network. John and the Pouzin Society’s assertion is that it has been known for some time the Internet as it exists today is not properly built to scale and the backup plan is not going to be much better. Let’s examine why he and a few other folks in the know think this is true.
The first problem is one of multihoming. More and more organizations are connecting multiple service providers for redundancy. When you do this by multihoming with BGP, a client that has a network block (let’s say a /24) is announcing a /24 to the service provider they probably got it from (Provider A) not to mention the other service provider they are utilizing (Provider B). So that route has gone from being an aggregated route in a supernet that Provider A is allocated to being the supernet, a /24 from Provider A and a /24 from Provider B. It went from 1 route to 3 routes in the global BGP table. Then consider clients utilizing multiple /24s and more than two service providers for their fancy failover/anycasting tricks and you can see, these scenarios make for rapid growth of the BGP table. If the rate at which it grows continues, there will be a time where the sheer size outstrips routing hardware capacities. People have been screaming about this one for a long time. Big deal, right? It is a fairly big deal, but we have farther to go and many much more interesting challenges.
Let’s examine the issue of IPv6. IPv6 is being banked on to fix the inevitble fact that we are of running out of IPv4 addresses pretty soon. Well, the IPv6 folks didn’t think about a fundamental, yet important detail regarding IPv6. An IPv6 prefix consumes 4x more memory in a router than an IPv4 prefix and requires more computing power for convergence of routes. Compound that with the increasing popularity of multihoming (that’s not going away), the crazy ways people are going to start chopping up IPv4 when the RIRs run out of blocks and the rapid uptake of IPv6 as IPv4 dries up then you have an interesting set of circumstances. Yet another taxing effect on our fair Internet routers. Getting scared yet? Well that’s not all, there’s more.
How about that fancy LISP protocol we talked about before? Isn’t the separation of location and identification going to make the Internet scale better and save us all? Not according to Dave Meyer @ Cisco. In his Internet-Draft titled Loc/ID Split Implications, Dave details circumstances in which determining the “liveness” of a locator path (i.e. knowing that a path to a host is still there) has not been properly accounted for. The implied conclusion is path discovery is required to know if a host is available when using loc/id split. Dave and others assert that this practice will not scale. Another conclusion is made based on this discovery that any network utilizing today’s protocol restraints requires routing/addressing over an interface will not scale.
Therefore, because IPv4 and IPv6 require an addressed interface for routing, they will not scale.
Does this mean the end of the Internet forever? Probably not, just the one we know today. I think their efforts are a call of action to say “Look dudes, our plan B to replace what we have is broken. We need to get to a plan C and fast.”
I’m sure I’m oversimplifying. The goal of this post was to get the concept out for debate. Feel free to leave a comment with your thoughts. To read more and make your own conclusions, there are some great resources on the Pouzin Society’s website at http://www.pouzinsociety.org
Recent Comments