Congestion Lessons from the World Cup

While soccer fans across the globe have been captivated with the excitement of the World Cup during the last month, vendors of DPI-based technologies have been casting their focus there as well. I recently came across a number of different vendor reports and musings (from Arbor, Allot, and Sandvine) about the impact of World Cup viewing on the Internet, and they got me thinking about the relationship between changes in Internet user behavior and application-specific congestion management.

Arbor, for example, showed that during certain World Cup games, global inter-domain video traffic using Adobe Flash (the most prominent technology used to deliver web video today) was showing 50-100% gains over an average non-World Cup day. While data about last-mile Internet connections is mostly absent from Arbor’s analysis, the fact that products like Arbor’s are picking out Flash traffic made me wonder whether last-mile ISPs might be singling out Flash traffic and applying particular congestion management policies to it (prioritizing it, for instance). While this might seem expedient in the short run (especially during the World Cup), Internet video technologies are largely in flux, in particular given the burgeoning maturity of HTML5, the evolving web standard that includes video support. If ISPs target their traffic management at Flash, what does that mean for new video technologies like those that HTML5 supports? If Flash video is given priority during high-volume times, newer video technologies may be at a disadvantage before they ever have the opportunity to see wide adoption.

But even if ISPs and DPI vendors can see new applications and content formats like HTML5 video on the horizon, using application- or content-specific congestion management would still require them to constantly upgrade the systems they use to identify and manage those particular kinds of traffic. Rather than playing a constant game of cat-and-mouse with applications developers, it seems as though ISPs would do well to focus their congestion management on the macro issue facing their networks – congestion – rather than the micro issue of which application or content delivery format happens to be popular at the moment. Congestion management solutions targeted at solving generic congestion problems would seem to make for better long-term investments.

Allot’s report points to another aspect of traffic management – its application to particular periods of the day. Looking at mobile Internet data, Allot noted large increases in Internet traffic (and World Cup viewing) during lunch time and in the afternoons as compared to normal usage. Allot also reported a huge uptick in streaming video on “post-match mornings” as fans flocked to their phones to view highlights from the previous night.

This data is pretty meaningful considering that some ISPs apply their congestion management policies during the same set of hours each day, usually in the evenings (for example, in the UK, Sky reserves the right to throttle streaming and peer-to-peer traffic of high-volume users between 5pm and 12am each day). Limiting congestion management to a rigid time window means that traffic spikes during other hours of the day, like those associated with World Cup viewing, may go unmitigated. Given the variability of bandwidth usage, sticking to a set traffic management schedule irrespective of the actual congestion status of the network also likely means that some users get throttled unnecessarily on evenings when the network is less congested. Managing traffic on a strict time schedule can thus be both over-inclusive and under-inclusive depending on how the network is being used.

Finally, Sandvine highlighted the rapid rise of Akamai’s HD content delivery server among the top 25 most-visited web sites on a particular Canadian ISP’s network. In just a matter of days, Akamai, which was serving video content for the CBC in Canada, surged from 19th to 2nd in the rankings, with 80% day-over-day increases in traffic volume. If ISPs were to have policies in place that prioritize or throttle traffic on the basis of its destination or origin, this kind of rapid change would be entirely unaccounted for. This again points to the long-term benefits of managing congestion on a generic basis, rather than singling out particular endpoints whose popularity may come and go.

Although the World Cup is obviously a unique event in terms of its global popularity, the impact that it had on video streaming is likely to be felt well into the future, and the infrastructure that was introduced to support it will certainly be reused for future events. As network usage patterns continue to evolve, flexibility in traffic management seems to be the key to addressing the three kinds of changes highlighted above: changes in application and content formats, changes in user schedules, and changes in network endpoints offering popular services. The World Cup experience points to the long-term value of approaching congestion in a generic way, agnostic to application, time of day, or network endpoint.

This blog post was first published by Alissa Cooper on her personal blog.