IPUrbia

Storm commented recently on an inherent dilemma of the Internet:

how do you do QoS? Once the demand for more exists, more will come. Whining about how the Internet’s going to get bogged down with P2P video seems a little short-sighted, and eerily similar to “you see, the Internet is a series of tubes”.

Again, however, I’m interested to see how this guy will juggle packet prioritization, keeping in mind latency issues and the inherent packet losses, etc.

I’ve been pondering over this issue lately, and I’m not sure that I have a good answer. I’m not one to scream “the sky is falling” (or, the Internet equivalent that the Internet is getting “full”) — I’m a firm believer in market forces and generally expect that, once demand gets high enough, supply will follow.

Still, a much more nuanced form of the question is in order: can the Internet function fully without end to end QoS (Quality of Service)? This question strikes at the heart of the Net Neutrality debate, and, for reasons that will soon be obvious, it is surprising that the telcos don’t trot this question out front and center. It’s a difficult question, although the answer is a tentative no, but it is a close call.

QoS, usually involving some form of packet prioritization, is considered to be a critical part of LAN design these days, especially in any networks carrying real time data such voice or video. For outside communications, application providers and consumers have developed several approaches to reliably delivering real time data over the Internet. A quick overview of these is important to the discussion, as the ultimate question revolves around whether these are sufficient long term methods or simply short term hacks.

• Last mile QoS: Many companies implement prioritization on their upstream link to the Internet. While this doesn’t actually affect what happens to the data once it is transversing the Internet, the general concept usually–but not always–works because the last mile link is often the most problematic since it is usually the smallest link on the connection. A loose analogy would be an ambulance that uses its sirens to bypass congestion on crowded city streets but then simply relies on the flow of traffic once it reaches the (generally) less crowded interstate.
• Private network QoS: While this is very obviously not the Internet, it must be mentioned because this is a huge market, especially for corporate customers who pay a premium for high quality private networks connecting their offices, generally using MPLS. This is sometimes done for security reasons, but is usually done to have transit links that can reliably deliver real timeVPN overcomes most security concerns.
• Application QoS: This is very much a misnomer (of my own creation), since it actually is not QoS at all! However, applications often take matters into their own hands and are simply written based on the assumption that their links are unreliable. Voice and Video over IP applications usually can make do with some degree of packet loss; video services usually “buffer” to give themselves a little bit of a safety net, and so forth.

data between offices: after all, a simple
While these methods don’t deliver end-to-end QoS, we have nevertheless been enjoying a fairly good voice and video experience on the “wild” Internet up to this point. It is important to point out that the last mile QoS generally suffices if the provider actually delivers the promised bandwidth, a novel idea these days in the consumer and SMB market. It might not be a bad idea if the status-quo persists, especially since it is much easier to have a content-neutral Internet in this environment.

Nevertheless, I do have to question whether this is practical long-term, especially as voice over IP takes hold. Simply put, is putting our entire telephone infrastructure on a “best effort” platform a good idea? Can the current design withstand a massive bandwidth surge? More importantly, can this architecture withstand a massive DoS attack by someone looking to take out the communications systems? What happens when 911 calls don’t work because of congestion? These are definitely red-herring arguments and a little bit of FUD, but have some measure of legitimacy, nevertheless.

I don’t feel that the solution has to involve a non-neutral Internet, although this argument could definitely be leveraged politically to justify one. A system that prioritizes ALL voice over ip traffic meets the mandates of good network design and while providing most of the benefits of “Net Neutrality”. Any other ideas?