Meta Ends Direct Peering with Deutsche Telekom

September 28, 2024

On Wednesday, Meta announced¹ it was ending its direct peering relationship with Deutsche Telekom (DT) due to the fee being charged for the connections. DT also had some strong words on their perspective of the situation.² This effectively means all DT customers will now have their traffic flow along a longer path through another internet service provider to reach Meta applications. It could mean slower or latent WhatsApp messaging, Instagram reels, or Facebook content. Not the end of the world for most people but it brings the architecture of the internet and the business decisions around it to news articles and public focus. It is also prompts a reexamination of peering agreements in the context of technology giants and how they work in building the internet.

To provide some background on why Meta made their decision we need to go back to 2010. DT and Meta, then still Facebook, struck a direct peering deal where DT set aside 24 private interconnection points with 50 ports and 5,000 Gbps at seven locations.³ All to handle traffic from DT’s customers towards Meta’s suite of applications: Facebook, Instagram, WhatsApp. Meta paid €5.8 million a year for these direct connections. All with the exclusive goal of making Facebook applications quicker for DT customers. Fast forward to 2020, and Meta asked for a 40% discount during contract renewal negotiations. DT countered with a 16% discount but then COVID-19 pandemic hit and it became less of a focus for both companies.

Later on in March 2021, DT had left the status quo intact and was offering to keep the ports running “for the benefit of consumers” while working out a new contract agreement with Meta. While this extended runway was provided to Meta, DT kept billing them for the original amount. However, Meta stopped paying the invoices and claimed these connections were now a “settlement-free peering.” They stated this kind of no fee direct peering setup was standard practice in the internet world. This is true for some peering arrangements but we’ll get into that later. This assertion by Meta was also proven by their experience globally with other telecom providers. DT was the unique exception that was charging transit fees.

DT obvious didn’t agree with this interpretation and in December 2022, the telecom giant sued Meta’s German subsidiary. The court ruled⁴ in DT’s favor, ordering Meta to pay €20 million. Under German contract law the continued use of the connections by Meta after stating the contract was terminated simply created a new contract. Thereafter, to avoid further fees under the new contract, Meta terminated these direct peering connections with DT.

The scope of impact is unclear for DT customers but we can already see the path changes network traffic towards Meta subnets. Based on 201 routes from DT’s looking glass tool towards Meta’s AS32934 I took an example route to show 3 of the currently available paths. 82 routes go through AS1299 Arelion, the artist formerly known as Telia. Only about 5 routes go through AS174, which is Cogent. The last 113 routes go through GTT’s AS3257.

These tier 1 service providers that are now handling this new transit traffic are large enough to handle it. However, DT claims Meta was sending approximately 3.5 TBps so the up tick will hardly go unnoticed by these 3 other providers. The technical aspects of this new routing situation are incredibly simple. All that happened for most traffic was an additional AS in the path. Further evidence is needed to see how much user latency is actually produced from it. But regardless of that outcome it allows us to move to the more complicated question of when or how settlement-free peering connections are normally accepted. Before that can be answered, I need to explain the context of DT’s perspective as a tier 1 provider and what that even means for peering agreements.

There is a 3-tier model classification system that organizes Internet Service Providers (ISPs) based on their size, reach, and interconnectivity. At the top of the hierarchy are tier 1 ISPs, which are the largest providers with global networks. These ISPs have extensive infrastructure that allows them to deliver internet traffic worldwide without paying other providers for transit. A defining feature of tier 1 ISPs is that they engage in settlement-free peering, where they exchange traffic with other tier 1 ISPs without any financial exchange. This arrangement exists because tier 1 providers have comparable network reach. With these connections among the tier 1 providers it allows them to reach the entire internet routing table and has the cool description of the global internet backbone. There are 14 tier 1 ISPs globally and recognizable examples for US residents include AT&T, Lumen, and Verizon. Deutsche Telekom is also a tier 1, which is an important point in order to understand their perspective on Meta’s claim.

Tier 2 ISPs operate on a smaller scale, often serving regional or national areas. They rely on tier 1 ISPs for access to parts of the global internet that their networks do not reach. While tier 2 ISPs may also engage in some peering with other ISPs, they often need to pay transit fees to tier 1 providers for broader internet access. Their peering agreements are typically limited to regional traffic or between similar-sized providers, which means they don’t have direct access to reach the full internet routing table. Comcast and Cogent are examples of tier 2 ISPs, as they operate large but regionally constrained networks that need transit services from tier 1 ISPs to achieve full connectivity to the internet.

At the base of the hierarchy, tier 3 ISPs are local providers that offer last-mile connectivity to homes and businesses within a specific geographic area. These ISPs do not have large networks and therefore rely entirely on purchasing internet transit from tier 2 or tier 1 ISPs. They do not typically engage in peering agreements due to their limited reach. Local internet providers serving specific towns or regions fall into this category.

Settlement-free peering is used when two ISPs, typically tier 1 providers, have similar network reach and traffic exchange patterns. Instead of sending network traffic through an intermediary provider, they agree to directly exchange data for free, as both parties benefit equally from the arrangement. This practice allows ISPs to reduce costs and provide faster routing for their customers. Peering relationships are based on mutual benefit, and for settlement-free peering to work, the traffic exchanged between networks needs to be relatively balanced. For tier 2 and tier 3 ISPs, peering is more limited or non-existent, because it is much more advantageous to simply pay to peer with a tier 1 that has much better global reach.

This description of settlement-free peering is entirely generalized and there are always exceptions or restrictions. The internet and the underlying architecture is a commercial venture and so these decisions are frequently business decisions rather than technical. Each provider has different rules and policies governing these decisions. Lumen has shared their Settlement-Free Peering Policy, which includes some steep hurdles like 500 Gbps per month and providing transit for 1,000 unique autonomous systems. Yet any of these specific points could most likely be ignored if it was in the specific business interest of Lumen as a company.

Based on the fact that DT is a tier 1 provider and now we know the general conditions for settlement-free peering does Meta’s claim hold any bearing? I would argue yes, but with the huge caveat which would require DT’s approval. I find it hilarious that Meta thought they could glide out of paying DT 5.8 million per year without them attempting to claw that all back. DT, after all, isn’t a charity. What incentive do they have to let a sizable cash flow suddenly stop? Any de-peering latency could easily be blamed on Meta, the foreign tech giant with, at most, a dubious impact on global society.

I can’t completely write off the situation purely on it being up to DT as a business. DT’s business decisions have come under increasing scrutiny as they have been exiting the industry norm. Barbara van Schewick has a striking summary⁵ of their attempts to extract revenue from both their subscribers and also business entities that pay for direct network access. Normally ISPs have provided neutral access for their subscribers to the entirety of the internet. But recent moves by DT and other providers require business entities to pay up or have network slowness for their service.

From a technical perspective, the use of a settlement-free peering connection would be helpful to both companies. The massive 3.5 TBps being consumed by DT’s users could be served much better directly between the networks rather than through the intermediary providers noted above. For fixed internet customers in Germany, DT is a provider for 40% of them.⁶ DT is ignoring the fact that their users are also their customers and are choosing willingly to request data from Meta. Where is their obligation as a provider to their users in order to decrease latency and improve the experience of being online no matter where that destination might be?