Comcast's Congestion Catch-22

Comcast’s Congestion Management, Take Two

Comcast has been very transparent about its second take on a congestion management system, filing a detailed description of the platform with the FCC and describing it in some detail on its customer “network management” page. To make the approach “protocol-agnostic,” the system does not examine which protocols or applications are consuming bandwidth. Rather, it examines which individual users are consuming excess bandwidth (defined as more than 70% of the subscriber’s provisioned service level -- i.e., the service level they are paying for) for a specific period of time (which Comcast measures in 15-minute increments). In essence, if a portion of its network reaches a congestion level, Comcast limits its heaviest users for a short period of time until that congestion goes away.

The implementation details reveal that the solution relies on three servers. The first is an IPDR (Internet Protocol Detail Record) server that collects usage information from the cable modem termination system (CMTS) in the Comcast network to track how many bytes a subscriber is consuming (Comcast has yet to publicly name its IPDR vendor). A second server from vendor Sandvine uses the Simple Network Management Protocol (SNMP) to measure CMTS port utilization and detect when a port is nearing a congestion state. When that happens, the Sandvine server asks the IPDR server for a list of users over their allowed consumption state, and if there are any (there may not be), it notifies a policy server from vendor Camiant. That server then compares the individual usage data to the network rules (individual user, over 70& capacity, for 15 minutes) and signals the CMTS to lower throughput for any over-consuming users. Comcast has only two quality of service distinctions: Priority Best Effort, which is used to deliver packets when there is no congestion, and Best Effort, which is used to manage the packets of individual users when they are over the consumption limits in times of congestion. BE-tagged traffic still gets through, but at a lower priority level.

A few additional details of the implementation are worth noting. In this deployment, Sandvine’s deep packet inspection (DPI) box isn’t really inspecting packets at all. It is not sitting “in-line” of the packet flow, and its capabilities to peer into the contents of packets aren’t even turned on, according to one source with knowledge of the system. That, of course, is part of the point – the FCC demanded that Comcast not discriminate against certain protocols, so it is not. The Sandvine network element is simply monitoring network usage levels, not inspecting packets.

“The product Comcast bought from Sandvine for this system reads and analyzes IPDR records generated by the CMTS, which includes byte counts per DOCSIS [Data Over Cable Service Interface Specification] flow,” said another source with knowledge of the deployment. “It is not their DPI box but essentially an ‘analyzer’ that looks at these usage counts per flow and measures how much is used over time. It isn’t capable of doing DPI and is completely protocol-agnostic.”

Also the policy that the Camiant system is enforcing is an extremely simple one – is there congestion, and if so, who is causing it? Policy servers from Camiant and others have the ability to do much more fine-grained control as well as link to real-time charging and other systems to administer a range of more fine-grained actions to help manage and rationalize network usage and congestion.