When the latest Akamai's State of the Internet was released earlier this month, I reviewed it as I always do, and something in the data intrigued me. It looked like in a large majority of geographies, in the last or last two quarters, average speeds had gone down.
The average speed measurement from Akamai is an interesting one, because as far as I can tell, it's as close as there is to a quality of experience measurement. It's not a test designed to test the speed of the line, like M-Lab or Speedtest. It's an actual measurement of the speed hitting the Akamai server for content replicated there by individual end-users. Of course, it's hard to know exactly what that content is although one can safely suppose that it is legal content (since companies pay for content to be replicated on Akamai's CDN), video or transaction heavy (since the only things you'd want consumers to get to real fast nowadays are these) and that it excludes (but may be parasited by) IPTV traffic.
I wasn't sure if this was a data glitch or a real dip, so I mapped the whole history of Akamai's SOTI data in an excel spreadhseet and started looking at the data in more detail. Here is the core findings:
- 36 out of 49 countries have average speeds in Q4 2011 equal or lower to Q2 2011. Some went up the one to last quarter and down more the last quarter. They are split as follows: Europe 20 of 23 countries, APAC 4 of 10, Middle East 6 of 7, Latin America 4 of 7 and North America 2 of 2.
- 21 out of 49 countries have average speeds that went down for these last two quarters. They are split as follows: Europe 14 of 23, APAC 2 of 10, Middle East 2 of 7, Latin America 2 of 7, North America 1 of 2.
In the past, there have been ups and downs in individual countries in the measurement, but only once (in Q2 2009) was there a similar drop accross the board when 34 countries out of 45 (Akamai didn't track Eastern Europe back then) were below their average speeds of two quarters before. Since then, the highest number of countries dipping was 13 in Q2 2010.
A mapping of all the countries is illegible, so I've rebuilt regional averages. Here is the graph I used in my Freedom to Connect presentation last week:
I'm trying to be cautious about this data, and definitely think it's too early to call it a trend, but I can't dismiss it either, it's too massive. I tried to figure out why these average speeds might be going down. The possible explanations I came to were the following:
- Increased use of mobile broadband: this was my initial thought but then I scrutinized the methodology section in the latest Akamai SOTI and noted that mobile Autonomous Systems are excluded from the broadband analysis. So that's not it, although see below for more on that.
- Akamai Methodology Change: when you have an accross the board drop in a dataset such as this one, you immediatly think that something in the way the data was measured has been changed. Akamai points to no such change however, and since they've been consistently producing great data, I'm inclined to trust them. The fact that many (but not all) countries have two consecutive drops also suggests that it's not a methodology issue.
- Lowering of Speed Package Subscriptions: in a context of international recession, it could be argued that consumers may be downgrading their broadband packages to save money, trading comfort for extra disposable income. That could explain an accross the board drop since the recession is largely global. However, we would have overwhelming evidence of that in various operators' financial performance and annual reports. And we don't. In fact, the message from the operators is the opposite: despite the crisis, people aren't downgrading or cancelling their subs much if at all.
- Actual degradation of the quality of experience in wireline broadband: that leaves us with the most likely explanation at this stage of the analysis, which is that the quality of experience is actually degrading. That could be explained by both an increase in devices connecting on the same line in the home and the increase in size and usage of online content, not compensated by a commensurate increase in broadband speeds offered.
Another possible explanation that has been pointed out to me by Robert Kenny is that since Akamai processes Autonomous Systems that are exclusively mobile separately, there may be ASs that are mixing fixed and mobile traffic and therefore influence the data. The exact paragraph in the SOTI report says:
Data is included only for networks where Akamai believes that the entire Autonomous System (AS) is mobile – that is, if a network provider mixes traffic from fixed/wireline (DSL, cable, etc.) connections with traffic from mobile connections on a single network identifier, that AS was not included in the source data set.
I don't know enough about Autonomous Systems to assess whether this could be a frequent occurence or not, mudding the data we're looking at. What would puzzle me if this was part of the explanation is "why now" ? Mobile broadband has been around for a few years now and should have weighed on the average speeds earlier if that was the case ? Or has something happened in the last two quarters in the mobile industry that I'm missing and could explain this ?
As I said earlier, it's too early to draw conclusions, but this is something that definitely needs to be tracked in the next two quarters, especially to see if the trend continues. The early numbers seem particularly worrying for Europe and North America, not so much for APAC.
If you h ave some insight as to why this might be happening, please feel free to chime in, either in comments or by email.
