An email I received this afternoon contained a forwarded link to an article entitled “Conflating broadband speed with Internet speed is misleading“. The article makes a valid point that access capacity (“Broadband speed”) isn’t the same thing as end-to-end throughput (“Internet speed”). Clearly this difference is valuable for consumers to understand, and is a critically important distinction in the Network Neutrality debate.

Which is why I’m disappointed in the article; sadly, it oversimplifies the issue to the point of covering up critical details.  The comparison to fax technology is imperfect, maybe even flawed. It conjures an incorrect conclusion in the mind of a reader. And the material result of this is to avoid a discussion of provider responsibility for effective bandwidth.

To be clear, end-to-end throughput across a network is affected by everything in between the two hosts (computers) that are communicating. It is affected by the equipment, configurations, and interconnects. It is also affected by the capability of the transport protocols, round-trip latency, packet overhead, and more. In this regard, the article is correct to say that effective bandwidth shouldn’t be compared directly to broadband access capacity. But likewise, to compare the effective bandwidth to the coding rates of fax machines is a vast oversimplification.

Looking at the factors that might affect end-to-end performance, a number of those are directly in the hands of the network provider.  The access link (i.e. broadband connection) to the customer is just the first component.  It terminates in an edge / aggregation network that is probably oversubscribed.  The edge networks may be interconnected across a backbone, with its own bandwidth constraints and physical distance inefficiency.  And Internet connectivity, to the backbone or to the edge network directly, is enabled by a number of peering and/or transit connections that are not necessarily equal.  This isn’t even considering the possibility of NAT, security, or bandwidth management devices that might constrain effective throughput.

When all is accounted for, there may be a considerable oversubscription rate.  Not that oversubscription is inherently bad; most users aren’t using 100% of their bandwidth at the same moment in time, allowing the provider to time-multiplex their users without causing negative performance.  And this oversubscription allows the provider to make money in an otherwise low-margin business.  But because it’s hard to determine how oversubscribed a provider is, they’re often tempted to push costs lower by oversubscribing more.  (Which is evident when providers get irritated by increasing usage, such as P2P traffic, by their customers.)  Further, the Internet transit connections might be acquired on-the-cheap, offering lower quality network paths (read: more oversubscription, more latency).  And the effect of these choices directly accrues against end-to-end performance.

Now, to be clear, I’m not advocating regulation of how service providers build their networks.  It should be up to each business to determine for themselves what is an effective network topology, interconnect strategy, oversubscription rates, etc.  But to focus the entire network debate on the access connections while ignoring the complex network that interconnects those to the Internet is misleading.