Several years back, I read an industry analyst’s paper titled “The End Draws Nigh for Settlement-Free Peering: Migration to paid peering agreements are now underway”. The report questioned the earnings outlook for companies with imbalanced traffic ratios -- in other words, any business sending more traffic than it receives -- and implied that the free exchange of traffic was going the way of the dinosaur. The report made more outlandish claims still, such as alluding to mysterious “major changes” coming for internet routing, with “negative implications for Cisco”...
Though the sky did not fall in the years to follow, a frequent conversation topic among backbone networking geeks remains peering, defined by Wikipedia succinctly as the “voluntary interconnection of administratively separate Internet networks for the purpose of exchanging traffic between the users of each network.” We ask ourselves time and again, with general-purpose Internet transit services so cheap, and widely available, does peering still make sense? Just last month, I was invited to speak at the Telx Marketplace Live event, among a panel of industry colleagues being asked the same question: is peering dead?
My answer is, most definitely not. But the rationale, process, and interests of peering partners have undergone major changes over the past decade.
Without a doubt, one trend I have seen is the formalization of interconnection. Back in the day (let’s call this early 2000’s for my generation), many peering agreements -- including those between major corporate entities exchanging significant amounts of traffic -- were literally made over a handshake, at a technical conference (or neighboring bar…) where network operators met.
These same agreements are now often filtered through formalized review boards and processes, and are accompanied by a legally binding contract. Due to M&A, some large networks have become even larger, and use this position to exert greater control over access to their monopoly/duopoly broadband subscribers.
As scrappy “little” broadband providers, SBC and Adelphia operated networks with good interconnection policies, where backbones and CDNs were able to exchange traffic effortlessly, and at no charge. SBC subsequently morphed into the “new” AT&T, which operates a rather congested network for those not willing to pay Ma Bell (sometimes onerous) access charges; likewise with the transition of Adelphia to Time Warner Cable. It’s also important to consider that not all traffic is subject to these obstacles -- many hundreds or thousands of broadband access networks, including some of the largest, "Get It" and still exchange traffic with content providers openly.
At the same time, new opportunities to connect networks together have emerged, thanks to a hugely interconnected hosting ecosystem. In 2005, launching a hosted presence on the Internet was relatively simple, even at scale: you’d choose a provider, and spin up some server(s) in a datacenter. In contrast, in 2015, the same public-facing web property might purchase a litany of services, for example:
- static assets cached on a CDN, such as a Limelight or Akamai
- object storage on Amazon’s S3 service
- bare metal hosting on Packet hardware
- additional or DR bare metal hosting on IBM’s SoftLayer
- development VMs on DigitalOcean and Amazon
- advertising brokerage
- yet more cloud services on Amazon, Microsoft Azure, and/or Google Compute Engine
The list goes on and on, requiring robust, highly available, connectivity between components. Here’s a hint: “over the public internet” isn’t the answer when you have tens of gigabits of business-critical data involved. In some cases, these new use cases for interconnection run circles around any doors closed by tightening peering policies, thus resulting in networks of today with greater net peerable traffic.
Here at Packet, we understand the interconnectedness of our industry, and its importance to our clients. Though we’ve just opened our doors to paying customers, we are already operating a fiber ring connecting our New York metro hosting facility to area carrier hotel locations, where we connect to to other broadband and content networks at major Internet Exchange Points like NYIIX and DE-CIX New York.
While the ink is still drying on these contracts, we’re already plotting our world dominance, beginning with network expansion to Ashburn, VA, the midwest, and the west coast. We’ve also made the decision to publish data on how we connect with various ecosystem partners, beginning with the next revision of our web site, to provide needed transparency through the murky waters of “who’s connected to who”.
By now one might ask, as an up-and-coming network operator, is peering right for you? Before arriving at an answer, it is important to take a step back, and look at your organization’s motivations to interconnect. No two organizations share precisely the same drivers, however there are numerous common themes -- in addition to the ones I just detailed, you might want to: add burst capacity (for absorbing Denial of Service attacks, or activities generating a temporary influx in traffic); obtain direct business relationships and operational contacts at networks (for troubleshooting performance or security issues); reduce expense by eliminating a middleman; and/or improve KPIs around network performance, also by reducing a potentially-congested middleman -- just to name a few.
Peering is by no means a walk in in the park, and absolutely requires an individual inside any organization to champion the concept, who is knowledgeable both technically and financially. Whatever the reasons, peering is certainly flourishing today, and any reports of its death have been greatly overexaggerated.