Ever since I started playing with networks, I’ve particularly geeked out on the value added by routing. I guess it is the same pleasure some get from train sets – the dopamine hit of seeing traffic come in from one direction and be intelligently split out to multiple directions. The things we did in IP routing ten or fifteen years ago are in wider use today, but still futuristic for most. Selective route download so we can use layer 3 switches as high density, low latency routers, for one. Or making containers first class citizens with the ability to push their network information right to the edge, including required firewall rules. Or massive use of anycast in internal services for high availability and locality. There’s more excitement coming to the IP network which, in our world, is overdue for another leap forward but I’ll come back to that another time. What I want to discuss today is voice routing.
Like IP, traffic going from network A to network B directly, and avoiding going through network C, D and E along the way is desirable. Back to a full-sized the train analogy: do you want a direct train from London to Glasgow, or do you want to change three times in Bristol, Exeter then Birmingham? Or do you want to go from London to Glasgow without changing, but stopping at every local station? It is obvious that the direct route is quicker, less prone to hold-ups and gives a better quality of journey. It is also obvious that stopping at every local station is going to take longer, and being smacked on the back of the head by someone else’s holdall each time the carriage population changes is going to wear thin.
It is exactly the same in IP, although in IP the journey time is milliseconds and most people don’t notice until something goes wrong, which it is more likely to do with an indirect route. In voice, which has to be real-time, latency matters but in milliseconds or tens of milliseconds is imperceptible – you need 160ms or more before the conversation starts to feel weird. What really buggers things up for voice is lost packets (think cancelled connecting trains) or jitter – variance in the latency. You can think of it as stop-start, a bit like the train stopping at every local station except it isn’t really. If two packets, or slices of a conversation, take different journey times, they may arrive out of sequence necessitating one of them be discarded. There’s things we can do to compensate, like delaying everything a few milliseconds, i.e. buffering, but that naturally affects latency. Whichever way you skin it, you want lowest latency, and the lowest jitter, i.e. most consistent latency. That is most certain from a direct route, A to B.
Then there’s security. In our train example, if people are coming and going every stop, or you’re having to change at other stations, is it more or less likely you’ll have your pocket picked or your phone swiped off a table than if you were never changing and going directly from London to Glasgow with an unchanging cohort of people? And so it is with IP or voice – the more men-in-the-middle you insert (crime doesn’t seem to have caught up with equality yet), the more likelihood there is of your call being intercepted, eavesdropped or misappropriated. If you don’t know who is in the middle, you don’t know what crappy questionable network gear they’re using and what level of Chinese State intercept is possible, let alone what other poor standards they are securing their network to. Again, direct is better, and that is before I get distracted with a rant about encryption. Even with encryption though, it is generally point to point in voice so you have to have fewer points in the path.
Without doubt, the most voice-connected network in the UK is BT, but their network is large and inefficient. For example, the locations of their SBCs are a matter of “national security” they claim so the notion of directly interconnecting with a node on their network, or even close to a node on their network, is unachievable. Instead you connect in a location, they backhaul it over their MPLS network to a secret location you’re not allowed to know, and which has no bearing on where call traffic is actually flowing. So, it is perfectly feasible to interconnect in London, to an SBC which could be in Aberdeen, meaning any calls to London will go London<>Aberdeen<>London, and God knows where along the way. Yes, it is one network hop, but that network exhibits all the qualities of many network hops.
Several years ago, we embarked on our BTZero strategy, recognising that what we’d done in IP made most sense in voice. It was also driven by necessity as at the time we hadn’t found the special handshake to solicit an invite to the Secret Club and were in a weird place with TDM – capacity took 12-18 months to provision but BT had to approve how much we could have, ergo, we never had as much as we needed. BTZero saw us pursue more direct interconnects to give plentiful capacity, improve quality (the BT TDM network was prone to issues even back then) and improve economics. As you will have seen from our recent news with Three, we have now completed direct connectivity to anyone who is anyone (yes, and them!). That is all major fixed and mobile networks which represent nearly all our traffic, with the only traffic going to BT being traffic for BT itself. I would argue, and am willing to be proved wrong, that this depth of interconnect puts us in a league of our own.
It gets better though. Where we interconnect with another network, we do so in multiple places, fully meshed with our multiple Availability Zones. When you send traffic to an Availability Zone, we route it directly from there to the destination network but we (uniquely) employ latency based routing, i.e. we send it to the closest available SBC on their side. I defy anyone to have a shorter or more reliable path, especially coming from a paradigm of statically and manually configured magic-boxes.
So with more direct interconnects than almost anyone and traffic delivered directly to the destination over the shortest light-path available, I’d put it out there that we are more directly connected in the UK than anyone. That gives the best quality, greatest security and unbeatable economics. Is your traffic already on Simwood or taking a formerly State-run now privatised (but still State subsidised) route, or someone else reselling it and just adding a point of failure?
