The Battle Continues: Peering vs. Transit

Ask Dr. Peering

I see another wave of 10G port pricing drops across Europe. What does this do to The Business Case for Peering?

Ilsa Lund Laszlo

Ilsa - of all the blogs in all the blogosphere in all the web - thanks for walking into mine.

Decreasing the price of peering ports is a good thing for the peering community. When the cost of peering declines it is easier for networks to make the case to peering. Growing the peering community with more routes and more traffic peered across the exchange is all goodness.

One thing to remember though is that, in addition to the peering port fees, the total cost of peering includes colocation, equipment, transport to the IX among other things. While the cost of peering ports drop over time, the colocation fees have largely been flat and transport has dropped a bit in price. Collectively, these costs have not dropped as fast as the price of transit has dropped. So we continue with the Peering vs. Transit battle and “Does Peering Make Sense Anymore” question.

Analysis

As you point out, last week, the AMS-IX dropped the peering port fees by 50%. As a result, an AMS-IX 10G peering port is only 1000 Euros per month now.

What is the effect of this price drop? In the graph below we see the drop in peering port fees results in the a lower total cost of peering. As one peers more and more traffic away for free, the unit cost for this traffic exchange drops, resulting in the slightly lower curve shown below.

We also see in this graph a simultaneous drop in the price of transit. So we can see that there is the potential for the Peering-Break-Even (PBE) point to end up in about the same place when both of these things happen simultaneously. Thus dropping the price of 10G peering ports might demonstrate a value proposition that is just about treading water, keeping 10G peering about equal to the price of transit.

To illustrate with some sample numbers, let’s assume:

•10G Transport into an IXP:

5000 Euros per month

•Colocation fees:

1000 Euros per month

•Equipment amortized expense:

2000 Euros per month

•10G Peering Port:

2000 Euros per month 1000 Euros per month

_____________________________________________________________________

Total Cost of Peering:

10,000 Euros per month 9000 Euros per month

Under these assumptions, this 50% reduction in the cost of 10G peering ports leads to a decrease in the total cost of peering of only 10%. (The 10,000 Euros per month cost becomes 9000 Euros per month.)

Peering vs. Transit Calculation

If one is paying 2 Euros/Mbps in transit fees as an alternative to peering, the Peering Break Even Point is 9000 Euros/month divided by 2 Euros/Mbps ---> you see a Peering Break Even Point of 4.5Gbps. That is, if one can peer at least 4.5Gbps away for free, then it is provably financially rational to do so.

If one pays 1.50 Euros/Mbps then the Peering Break Even point is 6Gbps. One can save money if they can peer more than this amount of traffic, and they can grow this cost savings until about 7.5Gbps when they start hitting the bandwidth ceiling and have to upgrade.

Summary

Dropping the price of peering is the right thing to do for European IXPs. It benefits all participants, and it reflects the IXP pricing model in Europe where the price of peering ports approximates the cost of the peering ports. Contrast this with the U.S. where the cost of the peering ports is the price the market will bear. So this is a positive step but a small change in how peers realize the reduction in peering costs.

Observation

We also must acknowledge that this is the most that an IXP can do to reduce the total cost of peering. The IXPs don’t operate a transport network and don’t control the price of colocation. Therefore, the cost of peering is increasingly out of the IXPs control and therefore the cost savings realized by peering is increasingly out of the IXPs control. What we are seeing here is the best the IXPs can do.

It is therefore not surprising that most IXPs are now enabling reselling of their peering ports and why remote peering is starting to gain traction. Of the 630 AMS-IX customers, over 130 are remote peering. Remote peering is successful because it removes these relatively fixed colocation, equipment and transport costs from the peering equation and makes the total cost of peering a single value-added transport circuit. One company shared that remote peering moved the Peering Break-Even Point from 8Gbps (where it didn’t make sense to peer) down to 3Gbps, a point where it was easy for him to make the case to remote peer. Over time, I believe the peering market will bifurcate; the large traffic exchangers will build in to peer and the medium scale networks will peer remotely or not at all.

Other scenarios and assumptions. If the cost of transport is closer to say 2000 Euros per month for a 10G into an IXP, the drop from 2000 Euros to 1000 Euros is more significant as a percentage. Here a drop from 7000 Euros to 6000 Euros is a 14% drop in total peering costs. Again, good for the community but not a huge change in the total cost of peering. For those peering across a 100G peering port that is now half the cost, the financial benefits may be substantial, but I don’t currently have pricing estimates for the cost of transport used to fill the 100G port. This would be a good follow on analysis to do.

Summary

The peering risk concern as always comes when you project the continuing decline in the price of Internet Transit.

You will observe in the graph above, that transit drops in price at the same time as the peering port prices drop. The end result might be that the peering break even point remains the same - that is, you have to peer away for free maybe 5Gbps before you cover the cost of peering. From there until you upgrade the peering infrastructure, peering is provable cheaper than transit.

The problem is that there is little runway before one has to upgrade to a second 10G peering port. If you can peer between 5Gbps and 7.5Gbps then traditional peering makes sense financially.

So, it is great news when peering points drop their port fees, but remember that peering port costs are only a part of the cost of peering.