Speed data explained: Why Akamai seems so slow

Last week, we published the findings of the latest Akamai State of the Internet report, based on data gathered from their global server network. It showed that, in terms of average speed, Australia was near the bottom of the OECD table and getting worse.

Several in the industry suggested that at 4Mbps, the Akamai speed was unrealistically low. Perhaps, although with such a big sample (up to 200 trillion http/s requests over the course of three months), you'd have to wonder how they got it wrong. Akamai put paid to one theory that their speeds are artificially slowed by "rate limited streaming", saying none of the data comes from their native streaming networks. For the report, they treat all content the same, whether it's video, HTML, JavaScript, or software updates.

Yet other measures of speed tell a different story. Our own ZDNet broadband speed test shows an average speed of around 9Mbps in Australia, and Ookla, which gathers data from Speedtest.net, recorded speeds averaging 11.6 Mbps at the end of last year.

Jason Ashton, CEO of Big Air, said the Akamai results are puzzling. To him, the suggestion that less than 5 percent of users have speeds over 10Mbps just doesn't sit right. He pointed to claims from TPG stating, "More than 50 percent of TPG's ADSL2+ customers obtain connection speeds exceeding 10Mbps". Unless they're looking to cop another fine from the ACCC for misleading advertising, we'd have to assume some accuracy in that statement. And this heat map from iiNet, admittedly just for Sydney's inner suburbs, shows that back in 2008, there were few areas where ADSL2+ got below 6Mps.

An average of 4Mbps does seem low, and for speeds to be slowing seems a little unbelievable. It's possible that the delayed delivery of the NBN has slowed broadband investment and that is seeing us grow at a slower rate than elsewhere, but not go backwards, surely?

David Belson, Akamai's product line director for Custom Analytics & MCDN, admitted that there was a bit of reprocessing of historical data that could be to blame.

"As part of the redefinition of the broadband (2->4 Mbps) and high broadband (5->10 Mbps) thresholds, we had to go back and reprocess the historical data set with a new EdgeScape definition, which had the unfortunate impact of changing some of the historical data."

He also said that on delving into the data more deeply, they saw a significant decline in the number of unique IP addresses from two major network providers between Q1 and Q2 2012.

"This may be indicative of the deployment of Carrier Grade NAT (CGN) infrastructure (a/k/a Large Scale NAT) within the networks — such infrastructure could also potentially act as a bottleneck, lowering the calculated speeds."

But it doesn't explain the discrepancy in speeds between their data and other sources. In theory, theirs is more believable. The two speed tests are based on users opting to test their speed. It's a sample skewed by people concerned that their speeds are so low, excited that they're so high, or simply trying out a new technology.

David Belson said that connections to Akamai's residential network greatly outnumber business connections, which would also reduce average speeds for a given country. And there are other influences, too. Speed tests check the collection of a single file. Akamai measures the downloads of all content, including parallel requests. The average web page, he said, requires 90 requests for content. Much of these are small files that won't reach maximum throughput rates. That alone could be enough to suggest the discrepancy (and refute TPG's claims). It also makes the Akamai figures more meaningful, because it relates to real-life usage.

As to the original concern, that Australia is well down that pecking order, that still stands. Using data from Net Index, everyone's speeds are faster, but Australia still remains well down the list of OECD countries. We've moved from 7th from the bottom (using Akamai data) to 10th from the bottom, although the difference between us and the three below is marginal.

The conclusion seems to be: Speeds aren't slowing; what the real speed is depends on how you measure it, but Australia is still lagging well behind most developed nations whichever way you look at it.