Why buy your IP transit from a Tier 2 network?

By Simon Woodhead

We’ve banged the drum a fair bit about why you should buy your IP Transit from Simwood (contention free, low latency, largely optical etc.) but a common response, especially to our recent ‘FREE IP Transit‘ offer is ‘well I already buy from «insert Tier 1»’. So, in this post, I’d like to look at why that is probably the wrong approach to take, in all but one special circumstance.

By definition, a tier 1 network is one who doesn’t buy IP Transit from anyone; they don’t need to as they peer with other tier 1s. This means they can reach the entire Internet through their customers (or customers’ customers) or their peers, or their peers customers (or peers’ customers’ customers). Thus it defines the commercial aspects of their cost base and has no bearing on the quality or directness of the route. Back in the day, connecting to a tier 1 represented connecting to the Internet, but back in the day if a UK Internet user wanted to access a UK website, there’s a very good chance that traffic would traverse the Atlantic twice in order to go via a tier 1.

Thankfully, that is no longer the case thanks in large part to peering. By virtue of peering, the tier 1 carriers’ immediate customers (by definition tier 2 networks) connect to each other either with a direct fibre, or using a peering exchange such as LINX. Now, the majority of UK users accessing UK resources, stay in the UK.

Of course, the tier 1s have built their own networks out to global monsters, so in our UK scenario the trip across the pond might not be there should the access and content providers not be peering directly, as the tier 1 is likely local too. Even if that tier 1 is having to hand off traffic to another tier 1, they’re likely doing so locally now. However, that traffic is still traversing multiple networks, and those networks are quite big. That means lots of opportunity for congestion, latency, jitter, and all the bad things we dislike.

Those big networks are of course competitors as well and there are sometimes games and shenanigans that go on. In the latter days of our relationship with Level 3 a few years ago we found that any traffic destined for Cogent customers in the UK would have to traverse the US, despite both having very large UK networks. Some might think this is the IP Transit equivalent of the kind of dirty tricks we see in the voice world.

So, undeniably, the best user experience comes from networks heavily peering with each other over private interconnects or IXPs such as LINX. We’ve taken many customers down that path and are happy to do continue doing so.

However, peering can be political, and it can take a long time to build up relationships in multiple locations with many other networks. It isn’t an overnight thing either; connecting to the likes of LINX gives potential to peer with most ISPs in the world, including the tier 1s, it doesn’t mean they will. So what about the rest of the Internet?

Many think that a tier 1 is the answer here and it is fair to say there is a lot of vanity at play, in a very similar way to we see vanity in voice: “I’m big and clever because I resell the incumbent”. We obviously refute that in the voice world and argue the clueful deal with Simwood, and that reselling a former monopoly makes one neither big nor clever. Try telling that to a shiny suit stuck on transmit though. Alas, I digress: Can we draw parallels here in IP Transit?

Well, tier 2 networks are far more inclined to peer with each other. In fact, it is necessary to make the business model work. If you were to look at a graph of the Internet by perceived significance, whilst you’d have the tier 1s forming an inner core, they’d be surrounded by many, many more tier 2s. However, if you were to shade that diagram then for traffic you’d find the tier 1s were not the core and that the majority of traffic exists on and between the mass of tier 2 networks. As a great but long-gone ISP in the US, Mzima, said many years ago, it looks like a doughnut.

So your choice is to buy your transit off a tier 1, and be a few AS-hops from that mass of traffic, or to buy your transit from a tier 2 networks and stand a far higher chance of being directly connected to it. I’d argue the latter is better all day long! Here there is a very clear parallel with voice, hence our BT Zero strategy.

Of course, tier 2s, by definition, already buy tier 1 transit to complete access to the entire Internet where not reachable through their peers. On the Simwood network over 90% of traffic flows to peers, often over direct fibre connections, and less than 10% falls back on traditional tier 1s. Thus as a network with your own peering relationships, I’d argue Simwood represents a better home for the rest of your traffic than any tier 1. Furthermore, like other long-established tier 2s, we don’t just have a single tier 1 interconnect; we have many, across a number of tier 1s. So that means, even if your traffic cannot be completed directly to a peer, it is likely to go directly to the tier 1 which has the target customer (or customer’s customer), rather than going from tier 1 to tier 1 and so on.

The above applies in almost every case and we think any network buying transit from a single source that is a tier 1 would be better served doing so through a quality tier 2, and we’d naturally argue that in the UK, especially for voice and video traffic, we’re the best choice amongst them. However, in the interests of our usual honesty and transparency, there is one exception here. If your intention is to build a large tier 2 (or tier 1 for the fanciful!), you’ll need lots and lots of peering relationships to get there. When we were earlier in this journey we purchased transit from a very well peered tier 2 following all the above logic, but found it made a handful of networks less inclined to peer with us; their argument was that they already saw these routes over peering so didn’t need to. It is a flawed argument but an argument nevertheless. Those peculiar networks aside, which are likely irrelevant to you anyway, we cannot think of a single scenario where any network is better served buying from a tier 1 over a quality local tier 2.

So, about that FREE IP Transit?