During Tuesday’s FCC Technical Advisory Process Workshop (5+ hours of video), we heard testimony on Quality of Service (QoS) in terms of how it works and whether it’s viable in the first place.  Some of the testimony on QoS from academics like KC Claffy seemed to be out of step with reality.

The problem with Claffy’s testimony is that she cited one of the commonly misinterpreted papers on QoS in the world from two Internet2 researchers.  The Internet2 hasn’t exactly been a credible witness since their Vice President Gary R. Bachula testified to the U.S. Senate Commerce Committee that the solution to broadband and alternative to using QoS was to just give every home gigabit Ethernet access because it was a low cost solution that merely required $15 gigabit Ethernet adapters.  While that might work in a University setting, Ethernet is limited to 100 meters and totally inappropriate for broadband deployment where cable runs exceed 1000 meters.

The Internet2 paper from Benjamin Teitelbaum and Stanislav Shalunov which supposedly proves why QoS will probably never be needed or deployed doesn’t actually say what most people claim it says.  The reason for the confusion is that the conclusion of the Teitelbaum-Stanislav paper is inconsistent with the actual content of the paper and many of the assertions are just plain wrong, and we can see this from the following summary of the paper.  Note that comments in (parenthesis) are mine and not attributable to the paper.

• The Internet2 experimental network probably doesn’t need QoS because it’s extremely over provisioned due to the fact that it is extremely expensive and metered.
• Even so, the Internet2 might still require QoS because some applications can burst and cause congestion so QoS is needed to provide performance guarantees.
• QoS performance guarantees are needed for things like remote robotic surgery where we can’t afford to have even the slightest interruption.
• Price of capacity is low and falling (this completely ignores the fact that demand is increasing as much as the decline in prices which means the price of the needed capacity remains constant).
• Residential and rural access may be expensive and can be expensive (that is an admission that capacity is expensive for broadband).
• QoS equipment and “clueful” engineers are too expensive to be practical (yet Cisco testified in this FCC workshop that they can’t sell a networking product without QoS, and QoS is widely deployed in every FTTN VDSL2 implementation that runs IPTV)
• Interdomain (or end-to-end) QoS is “dubious” (This is not true because Interdomain QoS can and is being deployed through Interprovider MPLS VPNs).
• QoS over a few congested links often work well.  (Admits QoS can be practical in small targeted applications)

Once we look at what Teitelbaum and Stanislav actually said in the paper, we see that they admited QoS was necessary even in the most abundant capacity networks and practical in many other cases.  It is a far cry from the typical assertion that “Internet2 concluded QoS isn’t needed” or Claffy’s assertion that “Internet2 concluded QoS is a waste of money”.  Even when we ignore all the flawed assertions in the paper, it’s obvious that the conclusion in the Internet2 paper doesn’t match its own arguments.

How can it be a waste of money if the Cisco expert at the same FCC hearing stated that Cisco can’t sell a network device without QoS, and how can it be a waste of money if QoS is widely deployed on the last mile in the case of IPTV and widely deployed at the inter-domain level using Interprovider MPLS VPN?

The FCC was eager to know if inter-domain QoS was possible and whether it existed, but it does exist through Interprovider MPLS VPNs.  It just doesn’t get talked about a lot because the inter-domain peering agreements are NDA and even off limits to the government.

Even Skype’s Chief Technology Strategist Jonathan Rosenberg who gave a presentation later in the workshop admitted that every bit of QoS helps especially in the access networks and that end-to-end may not be necessary.  The only point of Rosenberg’s I would criticize is that Rosenberg seems to see everything as a software development issue when he said that he needed “APIs” (referring to Application Programming Interfaces) to be able to make Skype work with an ISP’s QoS.  But this is totally unnecessary because it’s easy to identify protocols by their behavior and just give them the necessary characteristics they need.  Cox for example already prioritizes Voice over IP (VoIP) applications including Skype, but this might get blocked if the FCC bans good network management practices as the current draft of the NPRM seems to do.  We hope the FCC makes the necessary exceptions to protect low bandwidth jitter sensitive applications and allow true neutrality on the network.