U. S. Broadband Speed Slightly Better in Latest Akamai Report

Akamai published the Q4 2012 edition of their State of the Internet report yesterday, and it’s pretty much as expected: the trends that have been evident since 2010 are continuing. Globally, Internet connections are growing incrementally faster, and we see this trend in the U. S., where connection speeds are increasing somewhat faster than they are in other high-speed nations.

•  The U. S. has picked up one place in the “Average Peak Connection Speed” that’s the best measurement of network capacity, rising from 14th to 13th as the measured peak connection speed increased from 29.6 Mbps to 31.5 Mbps.
• In terms of the “Average Connection Speed,” widely cited by analysts who don’t know what it means, the U. S. remains in 8th place world-wide. but we’re no longer tied for it as we were in the previous quarter; Sweden is right behind us on this one.
• In terms of “High Speed Broadband Adoption”, the proportion of IP addresses with an Average Connection Speed greater than 10 Mbps, we remain in 7th place, but now we’re tied with  Sweden.

Another notable trend is the continued increase in mobile traffic, about which Akamai’s partner Ericsson says: “the volume of mobile data traffic doubled from the fourth quarter of 2011 to the fourth quarter of 2012, and grew 28% between the third and fourth quarter of 2012.” This is a global figure.

The primary author of the Akamai report, David Belson, wrote a very helpful blog post last week explaining what the metrics mean, and I would commend it to any researcher who cares to understand what Akamai is measuring: it’s titled Clarifying State of the Internet Report Metrics. First, Belson explains the most widely misunderstood metric:

Average Connection Speed: As noted previously, this metric is calculated by taking an average of all of the connection speeds calculated during the quarter from the unique IP addresses determined to be in a specific country or U.S. state.  There are, however, a number of factors that can influence the average connection speed measurement:

• Parallel Requests: According to the latest figures (4/1/2013) available from the HTTPArchive project, the average Web page requires 90 requests for content.  As such, a browser will likely open multiple connections to Akamai for various pieces of content, such as the base HTML page, a shared Javascript, or one or more advertisements.  Given that these requests are all competing for a limited resource (the Internet connection’s bandwidth), each of these requests gets just a fraction of the overall resource, ultimately lowering the observed speed associated with each request.  (While these requests maybe sent over a single persistent connection from a browser to an Akamai server, they are each logged individually by Akamai.)
• Small Files: Many of the components that make up a modern Web page (such as images, CSS files, Javascript files, AJAX responses) are relatively small in comparison to software downloads and media files.  As such, the connection used to download these smaller files is often short-lived enough that it doesn’t exist long enough to reach maximum throughput rates.  (Think of it as being similar to driving on multiple short trips around your neighborhood on local roads vs. going on a multi-state roadtrip on the highway.)
• IP Address Sharing: In this increasingly hyperconnected world, a single user may have multiple Internet-connected devices, and an average household today certainly does, from smartphones, tablets, televisions, and gaming systems to thermostats and refrigerators.  If multiple devices behind a single Internet connection (unique IP address) are all consuming content simultaneously, then as highlighted above, each device will ultimately have access to just a fraction of the whole connection, ultimately lowering the overall speed calculated for each request to Akamai.  (Obviously, this is also true on a much larger scale for business connections, where requests from hundreds of users may appear to be coming from a single gateway IP address.)

One implication is that locales where highly concurrent browsers are widely used will show lower Average Connection Speeds even while users experience faster page loading. This is because newer browsers permit more connections. Here’s a comparison of some popular browsers:

Browser Name/Version	Connections Per Host	Total Connections
IE 9	6	35
IE 10	8	16
Chrome 25	6	9
Chrome 26	6	9
Firefox 20	6	15
Firefox 21	6	9

What’s the point of this? The number in the right-hand column is a divisor that cuts your total network capacity down to the unit that Akamai measures in “Average Connection Speed” for browser sessions. So if you have a 50 Mbps connection (nice round number that represents what I have) and you use Internet Explorer 9, Akamai will observe an Average Connection Speed (ACS) for your IP address as low as 1/35 of the pipe’s capacity when you’re loading some web pages. In fact, it can get even lower than that if you’re loading small files that don’t last long enough for TCP to graduate from Slow Start (all TCP connections begin at an artificially low rate) and also if you’re sharing the IP address with other users, as you typically do at home and at work.

While some people argue that ACS is representative of “user experience,” Akamai’s description doesn’t say that at all: the concurrent connections are “competing for the same resource” in order to make the page load faster, but the more connections, the lower the measured speed. Hence, user experience is actually measured better by Average Peak Connection Speed (APCS) because it’s generally going to reflect a low divisor, probably one, while a long file transfer is going on.

I’m saying two things:

1. Akamai doesn’t measure page load times, only the load times of specific elements in a web page.
2. Fast page loads will show slower Akamai ACS scores with some browsers, especially Internet Explorer.

As more people gravitate to Firefox and Chrome, the ACS number will become more meaningful.  Firefox has already followed Chrome’s lead in limiting the number of parallel transfers, and IE is becoming more reasonable, although it has a way to go yet as it allow nearly twice as many parallel requests for pages that get content from multiple hosts (not uncommon for pages with ads.)

The APCS score is easier to understand; here’s what Belson says about it:

Average Peak Connection Speed: As noted previously, to calculate this metric, an average is taken of only the highest connection speed calculated from each unique IP address determined to be in a specific country or U.S. state.  Within the report, we note that we believe that the average peak connection speed is more representative of Internet connection capacity.  By using the fastest measurement observed from each unique IP address, we are capturing just those connections that reached maximum throughput rates.  Often, though not always, these connections are associated with the download of larger files, such as desktop applications, games, or software updates.  Furthermore, these connections are likely occurring late at night, or during some other period of lower usage, so while closer the theoretical maximum capacity of the connection, the measurements are unlikely to be representative of true throughput during normal conditions.  Furthermore, as an average, it will also mitigate the impact of the extremes – while there may be users connecting at Gigabit speeds in a given country, there will also be users connecting at dial-up, satellite, or DSL speeds.

How much this figure degrades during periods of high usage depends in part on the network technology in use; SamKnows finds that fiber and cable don’t degrade much at all, while DSL can degrade as much as 15%.

I continue to believe that High Broadband Connectivity is Akamai’s single most important measurement for policy purposes. Here’s what it means per Belson:

High Broadband Connectivity: As noted within the report, this metric represents the percentage of connections to the Akamai platform, from a given country or U.S. state within a given quarter, which connected at speeds of 10 Mbps or more.  In order to “qualify” for inclusion in the rankings within the quarterly report, more than 25,000 unique IP addresses from a given country needed to make requests to Akamai at speeds above 10 Mbps in that quarter.  This threshold was instituted to prevent much smaller countries, with significantly fewer unique IP addresses making requests to Akamai, from artificially outranking larger countries with more Internet users.  Note that the High Broadband threshold speed was 5 Mbps from 2008-2011, and changed to 10 Mbps starting with the 1st Quarter, 2012 State of the Internet Report.

So this number tells you what portion of the user population is able to enjoy commonly used high-speed applications.

So that’s all you need to know in order to understand the Akamai reports.

Predictions

I’d like to add some predictions based on recent developments in broadband services. Comcast recently doubled the speeds of their mid-range broadband plans and cut the price of their highest speed plan in half, so the 20% or so of American consumers who use Comcast will clock in at higher rates in 2Q 2013, two SOTI’s from now. Another item is the ongoing high rate of conversion from long-loop ADSL to the fiber-backed short loop VDSL2+ services provided by AT&T and Century Link, and yet another is higher speed versions of Verizon FiOS. All of that is on the consumer side, but the connections that Akamai looks at include the commercial ones as well, and that’s the market where the impact of new fiber optic services is most pronounced. So we’re not done yet.

(Cross-posted from High Tech Forum.)