The Inquirer-Home

The Iphone isn't slowing down the Internet

Analysis Wasn't me guv
Mon Apr 26 2010, 15:13

DATA CONSUMPTION figures from the largest content distribution network on the Internet point the finger at Iphones for a significant decrease in average connection speeds in South Korea, the world's best Internet connected nation.

The report, produced by Akamai, claims to provide a "State of the Internet" view based on the firm's vast array of servers around the globe that are used to serve all sorts of data to consumers. Akamai runs the Internet's most popular content distribution network (CDN), offering up everything from AMD's drivers to songs over Apple's Itunes. It is the almost ubiquitous nature of Akamai that allows it, however inaccurately, to claim it is a yardstick of performance on the Internet.

Akamai's view of the Internet comes from serving petabytes of data every week, data which for the content provider is served in a cost effective manner by offloading it onto Akamai's network. So while firms such as AMD and Apple choose to run their websites on Akamai's network, it's far more likely that the majority of traffic on the CDN is large files, ones that mobile devices are less likely to consume.

Therefore when the company says that it believes the Iphone launch "was likely responsible for the significant drop in South Korea's average observed connection speed", it distorts reality and shifts blame to not only the wrong device but also to the wrong spectrum of technology.

South Korea, according to Akamai's own figures, has the highest average "connection speed" by some margin at 11.7Mbps. Put into perspective, that figure is 36 per cent higher than second placed Hong Kong. Brits will not be surprised to learn that the UK failed to make the top 10 with an average speed of just 3.7Mbps. To emphasise the 'quality' of South Korea's Internet infrastructure, it was able to achieve the 36 per cent lead over Hong Kong even with a 29 per cent year-on-year drop in observed speeds. It is that rather drastic drop that Akamai is attributing to the introduction of the Iphone in the country.

Laying the blame of decreasing average bandwidth at the foot of the Iphone is at best, unfair, and more accurately, wrong. While it's easy to flog the Iphone for its Mickey Mouse operating system, blaming the device that supports up to 7.2Mbps HSDPA connectivity is actually missing the point. Akamai, to its credit, said that the firm observed wireless bandwidth to be a "fraction" of that observed with wired connections, but it nonetheless decided to lump wireless connection speeds into the overall figures anyway. The fact of the matter is that it isn't wireless connection standards that pose a problem, rather the availability of service to meet the standard.

That smartphones need bandwidth of greater than 7.8Mbps is questionable, as according to the same report, "standard definition DVD quality" playback requires 5Mbps, well within the Iphone's HSDPA implementation. So while that might bring down the average bandwidth of a well connected country like South Korea, it might, if mobile operators got their act together, actually increase the average bandwidth in countries such as the UK.

Actually Akamai's data shows that UK mobile operators aren't doing too badly though recorded speeds varied wildly between the three operators surveyed. Akamai anonymised the data, presumably to the relief of the operators, nevertheless the worst performer managed just 1.3Mbps compared to the best, which was just under 2.8Mbps. Within the EU, providers in Austria, Poland, Russia and Italy managed to top 3.1Mbps, but again the report showed that performance between operators within the same country can vary wildly.

The biggest problem with Akamai's report is that taken at face value it may seem like a window into the performance of the Internet, when actually all it represents is a rather heavily tinted window into the performance of Akamai's network. A CDN isn't a typical network, as it has gross optimisation to push content near the 'edge' and avoid users having to traverse the 'inner network' within Akamai. While Akamai might, depending on device usage patterns, serve a significant percentage of Internet data in a particular session, its performance figures can hardly be classed as indicative of other online experiences, such as gaming or peer-to-peer (P2P) networks.

To classify Internet performance by average bandwidth can mask several underlying issues. Applications such as gaming require low latency and packet loss and while bandwidth can, up to a point, mitigate latency it has little to no bearing on packet loss. Given that most content consumed on mobile devices is currently likely to be in the order of several megabytes, dropped packets rather than the difference in bandwidth is likely to have more of an impact on perceived performance than any other factor.

Latency, though often quoted by gamers, simply doesn't play that much of a role in the majority of Internet applications. High latency to a web server results in a delay to either render the web page or start playing a video. Assuming the user has enough bandwidth to sustain the video and audio transmission, jitter, the term used to describe dropped frames, is caused by dropped packets not latency. There are numerous causes of dropped packets, nevertheless while bandwidth can be used as a crude measurement of size, packet loss and latency should be seen as a measure of quality, something that is clearly missing from this report.

As the Internet moves towards smarter, scalable ways of distributing content, companies such as Akamai will have to adapt. Their model of 'distributed' client-server content delivery is under threat from truly distributed swarming technologies such as Bittorrent. As the firm's network purpose is to serve data rather than acquire it from hosts, the report failed to show the level of asynchronicity between download and upload bandwidth. Given that technologies such as Bittorrent account for a significant percentage of Internet traffic, the lack of this data leaves a gaping hole in this, or any such similar report.

The validity of this report as a true state of the Internet is questionable. While it can be used to see the average download speeds within a country to Akamai's servers, it doesn't tell us about what, if any, traffic engineering or quality of service (QoS) policies are being employed by users' equipment and their Internet service providers (ISPs).

It is wholly possible that HTTP traffic, the type which typically take place between users and CDNs, is prioritised lower than other applications such as games and voice-over-IP applications. Akamai is unable to determine the contention between applications and devices within the user's local network. So in fact, what these figures are telling us is that users in South Korea are downloading data from Akamai's servers faster than anyone else.

Validity issues aside, Akamai's report does show some genuinely useful security statistics. According to the firm, 74 per cent of "attack traffic" was on port 445, the port used by the SMB protocol to share files and printers over Windows networks. Compare this with the second most trafficked port, 22, at just 5.2 per cent. Port 22 is used by the secure shell (SSH) protocol, which tends to suggest a fair number of brute force attempts at gaining access to Unix machines are occurring. According to the firm's data, more than 10,000 different ports were attacked, close to a three-fold increase in just three months.

As P2P technologies mature and become more adept at handling 'flash crowd' behaviour and live streaming, traditional CDNs such as Akamai and Limelight will have to adapt or face obsolescence. These quarterly reports by the firm do provide an interesting view of performance to its servers but to take them as a snapshot of Internet performance is fundamentally wrong. The idea that one firm, no matter how much data it handles, can provide an accurate "state of the Internet" and to lay the blame for a decrease in average connection speed at the door of the Iphone is misleading and fundamentally inaccurate.

That's not to say performance to Akamai's servers isn't important, given the data served, it is. However as P2P networks mature as a means of distributing data, the firm's role on the Internet and its view as a looking glass into the network may start to diminish. µ



Share this:

blog comments powered by Disqus
Subscribe to INQ newsletters

Sign up for INQbot – a weekly roundup of the best from the INQ

INQ Poll

Happy new year!

What tech are you most looking forward to in 2015