DSL Prime: Real Speed

Of particular interest: U.S. speeds are higher than usually estimated—some studies may fail to account for the high speed of cable broadband.

by Dave Burstein of DSL Prime and Future of TV and the Web Video Summit [January 8, 2008]

Free.fr: European Traffic Has More Upstream and P2P
Cisco CRS1 Extremely Powerful
The traffic load at Free.fr is heavy, because they allow every user to easily share home produced or captured videos with others on the system. Users are even encouraged to create their own "video channels," streaming at full definition to others on the network. Two Cisco CRS-1's handle the entire load with plenty of reserve capacity for growth. The folks at Free.fr give the CRS-1 a strong endorsement. The CRS-1's also offer IPv6, which is now an option for many of Free's customers.

Their traffic load is surprisingly different from typical U.S. deployment. In North America, Sandvine and others have reported a major shift from p2p to http traffic. P2p traffic has remained flat for several quarters, possibly because most people have built the music collection already. Http traffic, including streaming video, has grown rapidly. In particular, use of megaupload.com is enormous, as many people from bittorrent and other sharing.

90 percent of Free's traffic is peered, and only on 10 percent do they have to pay transit. This is typical in Europe, where Amsterdam, London and other cites have efficient peering exchanges. Many policy models assume most carriers pay for transit, and therefore include costs three or four times too high. Surprisingly little European p2p goes to the U.S., possibly because most file-sharing in Europe is eDonkey/eMule, little used in the states.

Downstream traffic at Free.fr outweighs upstream by 2:1, less than the typical North America ratio of 3:1 or higher. That's presumably a result of on-network video uploads.

U.S. Averages 4.8 meg Downstream, Europe 3.7
Mike Apgar's Ookla has performed over 200 million speed tests and the results should surprise those who claim the U.S. runs much more slowly.

North America 4427 kb/s down 890 kb/s up
Europe 3728 kb/s 847 kb/s
Australasia 3655 kb/s 465 kb/s

(There are possible issues with where his Asian/Korean servers are so I'm holding off on reporting that data.)

It appears to me that the relatively high downstream speeds in the U.S. are because cable modems are typically running at 6-10 megabits these days, while almost all U.S. DSL connections are less than 6 megabits. DSL dominates over cable almost everywhere but the U.S., meaning the faster speeds of cable modems leave many countries somewhat behind. Fiber is fastest.

From Ookla Speedtest

Downstream
Japan 11.5
Sweden 7.9
Latvia 6.7
Romania 6.2
Lithuania 5.5
Bulgaria 5.3
Netherlands 5.3
Germany 5.2
France 4.9
Russia 4.9
Norway 4.7

Upstream
Japan 5.2
Lithuania 3.4
Russia 3.1
Bulgaria 2.7
Sweden 2.7
Denmark 1.6
Czech Republic 1.3
Netherlands 1.3

The surprise here is the high speeds in Russia. The typical Russian deployment runs fiber to the basement and then copper (Ethernet or VDSL) to each apartment. That easily allows for 10 Mbps symmetric service. Building LANs may also explain some of the high figures in Eastern Europe, but I have no data.

The rest of Europe, Australia, and Canada were somewhat lower. Chile was highest in Latin America, at 1.5 down. All of Africa was less than a megabit down except Reunion and Morocco. The location of the test servers may affect the numbers for Latin America and Africa, but they seem generally accurate.

The advertised and "up to" speeds in these countries are considerably higher, and make inaccurate most of the previous international comparisons. I believe this is more accurate because of the 200 million tests, and many of the patterns are consistent with other data. People choosing to test their speeds are not a random sample, of course, and there are other issues in the data, so more results are welcome. But this is a valuable corrective

Our Time Warner cable modem is 2 or 3 times as fast as our Verizon DSL Our Verizon DSL has been rock solid at 3 Mbps down, 768 Kbps up. Our Time Warner cable modem is at 5 to 7 Mbps down or higher, 95 percent+ of the time, whether on speed test or FTP. I just ran a speed test at 9.9 Mbps and downloaded a Microsoft update at about 6. DSL in many countries can go "up to" 24 megabits, but loop lengths and line conditions reduce that.

I now believe DSL Prime's reporting has been off-base on this, overestimating the effective speeds on European networks. My working assumption (from Masayoshi Son, Australian and Canadian tests, and what I know about line lengths) was that many of the "up to 24" ran at 10 to 15 megabits; looking at this data, I suspect there are more in the 3 to 10 megabit range than I had expected. The lower European speeds are also due to many customers still on 1 to 2 Mbps DSL service even though their line supports more.

Most of the Comcast and Cablevision customers during this period were on a 6 to 10 megabit download service, and most of Time Warner on a 6 megabit tier. If there were frequent congestion problems, these speed test results would be lower. Not absolutely conclusive, but corresponds to my experience and most of what I hear. I suspect (but am not certain) that cable generally avoids downstream congestion. Those who do have problems are the ones complaining and more visible, but I believe leave a misleading impression of typical U.S.cable networks.

Thanks to Scott Wallstein of PFF for pointing me to this data, and Apgar, also the founder of the ISP Speakeasy, for putting it together.

Copyright 2008 Dave Burstein.
The DSL Prime Newsletter is reprinted with permission.

"The power of the printing press belongs solely to those who own the presses"
—A.J. Leibling

The Internet is the cheapest printing press ever invented.