Hi. Yea, it’s Bill Lumbergh. You mentioned something about changes in the peering ecosystem back in January. Yeaaa. I’m not seeing it. What do you mean?
-- Bill Lumbergh
Bill - Thanks for the note. I can now provide a little more color to my prediction about some of the emerging shifts in the peering ecosystem.
Consider the value of the IXP graph below. The value of the IXP to the peering population is shown on the Y-axis, and a variety of metrics are shown on the X-axis. The simplest measure of value is based on the volume of traffic that is “freely” peered across the IXP population, traffic that would otherwise go through a transit provider at a metered rate. Taking all peering costs into account, there is a point where the network operator can peer away enough traffic for free to cover the cost of participation. I call this the Peering Break Even Point, aka “the critical mass” point. Beyond this point, the IXP delivers a financially provable value proposition, and the population derives so much value, they are not inclined leave. In fact, they are deriving so much value that they might even pay more!
Importantly, note that the public peering value proposition trails off at the right side, reflective of the fact that even though the traffic volume continues to grow, the population growth slows.
Why? A few reasons:
Perhaps the cost for new peers to get to the IXP is too high, especially coming in from far away.
Perhaps the number of networks that will peer with them tops off. Eyeball networks are now a bit more selective.
Maybe the number of “interesting routes” stops to grow because the owners of these interesting routes are also available elsewhere.
Maybe the technology reaches its limit, although admittedly, the ethernet technology has scaled in Europe well past the multi-terabits per second range.
Some IXPs and networks have decided that they would rather migrate their high-volume strategic public peering partners to private peering using cross connects instead of continually testing the limits of metro ethernet.
Optics are cheap today, so even though folks complain about cross connect fees, they can push a lot of traffic over fiber these days with great reliability.
As a result of these and other factors, the industry continues to evolve toward a blend of public and private peering.
The Emergence of Market Peering
With the Open-IX initiative, we are also seeing a growth in the public peering options in the U.S. market. This is of course both good and bad. The good news is that the market’s “invisible hand” is now at work, providing choice to new participants as well as existing disgruntled participants.
Is the pain strong enough to cause participants to move from a traditional carrier-neutral colo into one that is blessed by the Open-IX certification? Perhaps. The peers that have bought into the Open-IX religion will extend or move, and this camp includes some of the largest traffic sources and sinks in the world. These movers contribute to the gravitational pull of the emerging certified exchanges.
However, for many networks, the pain of the traditional colocation centers is not bad enough to cause a move. These “set-it-and-forget-it” peers will not disrupt their operation to move out any time soon. They will simply let their contracts auto-renew indefinitely. Instead of moving out, these peers may instead tether across the market to pick up routes not available in the dominant colos.
[Disclosure: I work for one of these companies that provides tethering within and across markets.]
And so we are witnessing the peering ecosystem migrating from selecting a single IXP to build into, to selecting a market to build into, and then tethering. Let’s consider a particularly interesting example.
The Bad News: The Splintered Peering Markets
I took a snapshot of peeringDB to create the following Transit & Peering Services (TPS) report. What you can see from this report is the degree to which the New York / New Jersey market is fragmented.
(Apologies for the fine print. The TPS report for the New York market (244 peers!) was too big to fit, so I took a screenshot of the first 50 or so peers that are reachable in the New York / New Jersey market. Click on the TPS Report below to see it as a PDF.)
On the left side is the list of networks, and across the top are the IXPs they participate in. The marks inside the grid indicate peering inclinations (Open, Selective, Restrictive). The point is not the type of peering presence or inclination, but that peers are splintered across many IXPs in the market.
The other way to see the non-overlapping in the market is to look at the graphic that shows the colos and IXPs across New York. Which building to choose?
I think the physical presence justification is more related to private peering than public peering. Private peering is best done across fiber cross connects within a building, where public can be done from anywhere in the market, assuming tethering and local IXP ports are available in the building you choose.
Think Market not IXP
Which IXP to build into? The good news is that you don’t have to choose. You can reach all of these peers from any major colo in the market today. The transport between colocation centers across the market may included with your IXP port fees. Across IXPs can be accomplished by tethering or via cheap VLANs. With these two market transport options, the physical building and IXP selected becomes less important; you can reach all peering participants regardless of which building they chose.
The price points I have seen suggest that tethering between colos and IXPs that aren’t in the building (yet) remains less expensive than POP’ing a building, and the transport costs (if any) generally works out to be about the same price of Internet transit. So why not peer away that traffic directly across the entire market, instead of merely peering at a single IXP?
I hope this helps -