The Cogent-Sprint Depeering Incident (2008)

On October 28, 2008, Sprint (AS1239) severed its peering connection with Cogent Communications (AS174). Within hours, a portion of the internet effectively split in two: customers who were single-homed to Sprint could not reach customers who were single-homed to Cogent, and vice versa. The incident lasted approximately three days before a temporary peering session was re-established, but the fallout exposed deep structural vulnerabilities in how the internet's backbone is held together -- not by regulation or engineering mandate, but by voluntary business agreements between private companies.

The Players: Sprint and Cogent

Sprint (AS1239) was one of the original Tier 1 networks -- a backbone provider that could reach every destination on the internet through settlement-free peering alone, without purchasing transit from anyone. Sprint's network spanned the United States and connected to every other major backbone through direct peering links established over decades of internet growth.

Cogent Communications (AS174) was a different kind of Tier 1. Founded in 1999, Cogent built its business on aggressive pricing, offering bandwidth at rates that significantly undercut established carriers. Cogent attracted customers by being the cheapest option for high-bandwidth connectivity, particularly for hosting companies, data centers, and content providers who needed raw capacity more than premium service. By 2008, Cogent had grown large enough to claim Tier 1 status through an extensive web of peering agreements with other major backbones.

The two networks had maintained a settlement-free peering relationship for years. Under this arrangement, each network carried traffic destined for the other's customers at no charge -- a standard practice among networks of roughly equivalent size and traffic volume.

What Caused the Depeering

The precise terms of the dispute were never fully disclosed publicly, as peering agreements are private contracts. However, the core issue was a traffic ratio imbalance and disagreement over the terms of continued interconnection.

Sprint argued that the peering relationship had become unfavorable. In a typical settlement-free peering arrangement, both sides expect roughly balanced traffic flows -- each network sends about as much traffic as it receives. Cogent's customer base, heavy on content providers and hosting companies, meant that Cogent was sending substantially more traffic into Sprint's network than Sprint was sending into Cogent's. Sprint contended that this imbalance meant Cogent was effectively using Sprint's network as cheap transit rather than engaging in true settlement-free peering.

Cogent, for its part, viewed the traffic imbalance as a natural consequence of internet traffic patterns. Content flows from servers to eyeballs, and Cogent's customers hosted a disproportionate share of content. From Cogent's perspective, Sprint's end users were requesting this traffic -- Sprint's customers wanted to reach Cogent-hosted content, and Sprint was obligated to deliver it.

The dispute was fundamentally about money: Sprint wanted Cogent to pay for the traffic imbalance, either through a paid peering arrangement or by reducing the ratio. Cogent refused. Sprint terminated the peering session.

The Anatomy of a Partial Internet Partition

When Sprint withdrew its BGP routes from Cogent and stopped accepting Cogent's routes, the two networks could no longer directly exchange traffic. But this did not mean the entire internet broke. The impact depended on how each affected network was connected.

The internet's routing system works through a mesh of interconnections. Each autonomous system typically connects to multiple other networks for redundancy. When two Tier 1 networks disconnect, most traffic finds an alternative path through other networks. But some traffic cannot be rerouted:

• Single-homed Sprint customers -- networks that purchased transit exclusively from Sprint and had no other upstream provider lost all reachability to Cogent-only destinations. Sprint could no longer deliver their traffic to Cogent, and there was no alternate path.
• Single-homed Cogent customers -- the mirror image: networks relying solely on Cogent for transit could not reach Sprint-only destinations.
• Multi-homed customers -- networks with multiple upstream providers were unaffected. Their traffic to Sprint destinations flowed through Sprint (or another provider with a Sprint path), and their traffic to Cogent destinations flowed through Cogent (or another provider with a Cogent path).

The partition was partial, not total. The majority of the internet continued to function normally because most significant networks were multi-homed. But any host whose only path to the global routing table ran through one of the two depeered networks lost connectivity to hosts on the other side.

How BGP Propagated the Partition

Understanding why the partition occurred requires understanding how BGP route propagation works at the Tier 1 level.

In the pre-depeering state, Sprint and Cogent exchanged their full routing tables with each other. Sprint announced its customer routes to Cogent, and Cogent announced its customer routes to Sprint. Each network also received the other's routes indirectly through other Tier 1 peers, providing redundant paths.

When the peering link was severed, Sprint withdrew all the routes it had advertised to Cogent and discarded all routes received from Cogent. Cogent did the same in reverse. At this point, each network needed to find alternative paths to the other's customer routes through different Tier 1 networks.

For most Cogent customer routes, Sprint could reach them through other peers like Level 3 (AS3356) or NTT (AS2914) -- if those networks had learned the routes from Cogent directly or transitively. And they usually had, because Cogent peered with all other Tier 1 networks. The problem arose with routes that were only reachable through Cogent's direct customer cone and had no alternate propagation path.

A Pattern of Disputes: Cogent's Depeering History

The Sprint depeering was not an isolated event. Cogent had been involved in a series of peering disputes with other Tier 1 networks, forming a pattern that revealed fundamental tensions in the internet's economic structure.

Cogent vs Level 3 (2003, 2005)

In October 2003, Level 3 (AS3356) terminated its peering with Cogent, creating a similar partial partition. The dispute lasted several days and drew widespread media attention as the first major Tier 1 depeering to affect end users visibly. The networks eventually re-established peering, but the relationship remained contentious.

In 2005, another disruption occurred between the two. Level 3 again raised concerns about traffic ratios and the cost of carrying Cogent's heavily asymmetric traffic flows. The recurring nature of the dispute highlighted that peering agreements are not permanent -- they are ongoing business relationships subject to renegotiation and termination.

Cogent vs Telia (2008)

Earlier in 2008, before the Sprint incident, Telia (AS1299) also depeered Cogent. The Telia dispute followed the same pattern: disagreement over traffic ratios, demand for payment, refusal, and disconnection. Although the Telia depeering caused less visible disruption -- Telia's customer base was more heavily concentrated in Europe -- it demonstrated that multiple Tier 1 networks had reached similar conclusions about the sustainability of settlement-free peering with Cogent.

The repeated pattern raised an uncomfortable question: was Cogent's business model, built on undercutting competitors on price and driving traffic volumes up, fundamentally at odds with the settlement-free peering relationships that Tier 1 status depends on?

The Three Days: October 28-31, 2008

The Sprint-Cogent disconnection became effective on October 28, 2008. The immediate impact was detectable through BGP monitoring tools and route collectors. Within minutes, BGP sessions between the two networks went down, route withdrawals propagated across the internet, and the routing table reconverged -- minus the direct paths between Sprint and Cogent.

For affected users, the symptoms were confusing. Some websites loaded fine; others timed out. Some email was delivered; other messages vanished into the void. VPN connections to certain offices failed while other remote resources worked normally. The intermittent nature made diagnosis difficult for anyone who did not understand the underlying routing structure.

Network engineers who examined the situation could see the problem clearly in BGP data: routes to certain prefixes showed no valid path from Sprint's perspective, and the same was true in reverse from Cogent's vantage point. Traceroutes to affected destinations simply stopped at the boundary of the partitioned network.

On approximately October 31, Sprint and Cogent restored a temporary peering connection. The terms of the restoration were not publicly disclosed, but it was understood to be a temporary measure while longer-term negotiations continued. The pressure to restore connectivity came from both sides' customers, who were not willing to tolerate an indefinite partition of their internet access.

Why the Internet Did Not Completely Break

The fact that the Sprint-Cogent depeering caused only a partial partition rather than a catastrophic internet failure reveals important properties of the internet's design.

Multi-homing provides resilience. Any network that connected to both Sprint (or a Sprint customer) and Cogent (or a Cogent customer) acted as a bridge. Most significant internet services were multi-homed in 2008, meaning they had transit from multiple providers. Their traffic could always find a path around the gap.

The Tier 1 mesh provides alternate paths. Both Sprint and Cogent peered with other Tier 1 networks like Level 3, NTT, Telia, and Zayo (formerly AboveNet). For most destinations, traffic could take a longer path through one of these intermediaries. The partition only affected destinations that were exclusively reachable through one of the two disconnected networks with no alternate transit path.

Internet Exchange Points created additional connectivity. Networks that exchanged traffic at IXPs had direct peering relationships that bypassed the Sprint-Cogent link entirely, further limiting the partition's impact.

The Economics of Peering

The Sprint-Cogent dispute laid bare the economic tensions underlying internet interconnection. Peering is a voluntary arrangement, and the criteria for who peers with whom are entirely at the discretion of each network operator. There is no regulatory body that compels Tier 1 networks to interconnect, and no technical standard that mandates it.

Peering decisions typically consider several factors:

• Traffic ratio -- most peering policies require roughly balanced traffic exchange, often within a 2:1 or 3:1 ratio. Extreme imbalances suggest one network is effectively receiving free transit.
• Traffic volume -- networks typically set minimum traffic thresholds for peering. If the volume is too low, the administrative overhead of maintaining a peering session is not justified.
• Geographic scope -- Tier 1 networks generally expect peers to have a network footprint of comparable geographic scope, with interconnection points in multiple regions.
• Network capacity -- peers are expected to maintain sufficient capacity to handle traffic growth without congestion.

When these criteria are not met, the larger network may demand payment (converting the relationship from settlement-free peering to paid peering or transit), or simply terminate the connection. This is exactly what happened in the Sprint-Cogent case.

The challenge for the internet community is that these private business decisions can have public consequences. When two Tier 1 networks disconnect, their respective customers -- who have no seat at the negotiating table -- suffer the consequences.

Lessons for Network Architecture

The 2008 Sprint-Cogent depeering reinforced several critical lessons for anyone operating infrastructure on the internet.

Multi-homing Is Not Optional for Critical Services

Any organization that depends on internet connectivity for its business should have transit from at least two independent providers, preferably in different Tier 1 provider cones. Single-homed networks are vulnerable not only to depeering disputes but also to cable cuts, router failures, and provider outages. During the Sprint-Cogent partition, multi-homed networks experienced zero disruption.

Tier 1 Status Is Not Permanent

The concept of a Tier 1 network rests on the ability to reach every destination through peering alone. When a Tier 1 network loses a peering link, it may temporarily lose the ability to reach some destinations without transit, effectively losing its Tier 1 status in practice even if it retains it in name. The Sprint-Cogent dispute showed that Tier 1 status is a function of active business relationships, not a permanent designation.

The Internet Has No Central Authority for Interconnection

Unlike the telephone network, where regulators can mandate interconnection between carriers, the internet's interconnection is governed entirely by private contracts. This means the global routing table is, in a sense, the emergent result of thousands of bilateral agreements between independent networks. It works remarkably well most of the time, but depeering disputes reveal the fragility of this arrangement.

The Aftermath and Long-Term Impact

Following the temporary restoration of peering, Sprint and Cogent negotiated a longer-term arrangement, the details of which were not made public. The incident, combined with Cogent's earlier disputes with Level 3 and Telia, led to broader industry discussions about peering norms, dispute resolution, and the risks of the status quo.

Several trends accelerated in the years following the 2008 disputes:

• Growth of content delivery networks -- companies like Cloudflare (AS13335) and Akamai built massively distributed networks that peer directly with eyeball ISPs at hundreds of IXPs, reducing dependence on any single transit provider.
• Hyperscaler peering -- Google (AS15169), Meta (AS32934), and Amazon (AS16509) built global backbone networks and peered directly with ISPs worldwide, effectively bypassing the traditional Tier 1 transit hierarchy.
• IXP expansion -- the number and capacity of internet exchange points grew dramatically, providing more diverse interconnection options and reducing the impact of any single peering relationship.
• Increased multi-homing -- network operators became more aware of the risks of single-homed connectivity, and multi-homing became standard practice for any network with meaningful uptime requirements.

The traditional Tier 1 hierarchy has become less important as the internet has evolved. Content providers now carry a substantial fraction of all internet traffic on their own private backbones, peering directly with access ISPs and often bypassing transit providers entirely. The kind of partition that occurred in 2008 would have an even smaller impact today, though it would still affect single-homed networks on either side.

What BGP Data Reveals

You can still observe the structure of these networks today using BGP data. Looking up Cogent (AS174) in the looking glass reveals a massive network with thousands of announced prefixes and extensive peering. Sprint's original AS1239 has changed hands through mergers (Sprint was acquired by T-Mobile in 2020), but the legacy AS still appears in the routing table.

Examining the AS paths for any given destination illustrates the concept of multi-homing and path diversity that protects against depeering events. When you look up an IP address and see multiple routes with different AS paths in the route table, you are seeing the redundancy that prevents most depeering events from causing widespread outages.

Could It Happen Again?

Peering disputes have not disappeared. Networks still deactivate peering sessions over commercial disagreements. In recent years, disputes between content providers and access ISPs over interconnection terms (notably the Netflix-Comcast disputes of 2013-2014) have echoed the same themes: who should pay for traffic that flows asymmetrically, and what leverage does each party have?

The structural risk remains: the internet's global connectivity depends on voluntary business agreements between private companies. There is no guarantee that any two networks will maintain interconnection, and there is no regulatory backstop in most jurisdictions that would compel them to do so. The best defense remains the same one that worked in 2008: multi-homing, diverse peering, and not depending on any single provider for reachability.

Explore the Networks Involved

Use the looking glass to examine the autonomous systems at the center of this dispute and see how internet routing has evolved since 2008:

• AS174 -- Cogent Communications: look at its current peering and customer cone
• AS1239 -- Sprint (now part of T-Mobile): trace its routing table legacy
• AS3356 -- Lumen/Level 3: one of the Tier 1 networks that bridged the gap
• AS1299 -- Arelion (formerly Telia): another Tier 1 involved in Cogent disputes
• AS2914 -- NTT: a major transit backbone