We are interested in alternative ways of actively measuring the throughputs for countries of the world especially for developing countries. Part of this is to compare the results from direct measurements of throughput with those obtained from the PingER derived throughputs. with those obtained by the. We therefore contacted Ookla by email on November 27, 2007 and Mike Apgar and Doug Suttles of Ookla were kind enough to reply, answer questions, and on December 6, 2007 provide data suitable for importing into Excel (CSV format) and analyzing.
...
At the time of the data, Ookla speedtest.net had 193 servers in 68 countries. The locations of the servers are seen in the map to the left. The most servers/country were US (54), Australia (7), Russia(6), Brazil, Canada, Romania (5). A map of the number of servers / country is seen in the middle left below . Shown in the middle right is the number of servers/country for countries that had clients. Clients in countries without servers (red in the middle right map) will need to cross international boundaries from the server and so are likely to have poorer performance than clients with servers in their country. About 39% of the countries with clients also have one or more servers. The bar chart to the right shows the download speed per country together with trhe the number of servers per country.
...
The US has the most IP addresses (clients) = 5.7M, followed by the UK (2.5M), Canada(890K), Germany(84-K), Italy(835K), Australia(607K), Poland(556K), Mexico(481K), Russia(442K). A map of Clients/country is shown below. The lowest number of clients/country are French Polynesia(100), Greenland(102), Zambia(106), Burkina Faso(107), Haiti (108), MadagaskarMadagascar(118), Northern Mariana Islands(121), Belize(130), Mali(146), Mauritania(149). A frequency histogram (with logarithmic bin widths) of clients/country is shown to the right. the The relative flatness of the distribution (with logarithmic bins) indicates that the fall off in clients/country is exponential in nature.
...
The highest download speed is Japan=11Mbits/s, followed by Sweden, Latvia, Romania, Netherlands, Bulgaria, Singapore, Germany, France, US. Slowest download speed is Cameroon (199kbits/s) followed by Mauritius, Botswana, Angola, Madagascar, Nepal, French Polynesia, New Caledonia, Tanzania, Kenya (i.e. bottom 5 in Africa). Fastest upload is Japan (5.1Mbits/s) followed by Russia, Romania, Sweden, Bulgaria, Latvia, Hong Kong, Lithuania, S. Korea, Moldova. The large number of Eastern European countries is interesting and not fully understood. The slowest download speeds are Madagascar (70kbits/s), followed by Yemen, Angola, Cameroon, New Caledonia, Botswana, Uruguay, Nepal, Zambia, French Polynesia (276 kbits/s). 50% of the slowest 10 download speed countries are in Africa. The median upload speed (303kbis/s) is 1/3 of the median download speed. Currently the African representation by country is only 26 (out of I think 54). It will be interesting to see if this improves as more measurements are made.
A map of the Ookla download and upload throughputs measured for clients in the various countries of the world is shown below. The poor coverage for Sub-Saharan Africa (there was no data for countries in white) is seen as is the much better performance for developed regions such as N. America, Europe, Japan and Australaisia compared to developing regions. Also on the right we show a map of the ratios for upload/download.
...
A scatter plot of the Ookla upload versus download throughput speeds is seen below on the left. Also the ratio of upload/download is seen in the map to the right above. In all cases the download exceeds the upload speed. The best value for the ratio of up/down is 80% (Azerbaijan). The worst is 9% (Portugal), the median is 34% +- 17%. It is seen that several Eastern European countries such as Russia, Romania, Bulgaria, and Moldova have anomalously high upload throughput speeds. Looking at the map of Ookla servers and reviewing he the locations of Ookla servers in December 2007 there were 6 in Russia, 5 in Romania and 3 in Bulgaria. Presumably these servers can provide high throughput for clients. Several countries in Western Europe such as Portugal, Belgium, France and Germany have anomalously low upload throughput speeds. This may be the servers, or possibly the broadband is configured very asymmetrically. The figure below shows bar charts of the ratio of upload/download together with the download speeds, plus histograms of the ratio and download throughput.
...
Evaluation versus PingER Normalized Derived Throughput (NDT)
The results from Ookla are mainly oriented to broadband access to the residence and the performance of the local ISP. It also measures the true throughput. PingER on the other hand is more oriented to measurements beween academic and research sites and also derives the throughput using the Mathis formula (TCP throughput ~ 1460*8/(RTT*sqrt(loss)) where the result is in kbits/s if the RTT is in milliseconds. Though the Mathis formula is for classic versions of TCP (e.g. New Reno) as used in most off the shelf computers today, it assumes TCP loss (and, and TCP provokes loss to detect congestion) whereas PingER gets the loss from pings and thus only samples loss. For these reasons and others PingER is only a rough estimate of throughput.
To validate whether the order of magnitude of the throughputs measured by Ookla agreed with those derived from the PingER losses and Round Trip Time (RTT) measurements we plotted the Ookla speedtest download measurements versus the Normalized Derived throughputs (NDT) from Pinger. It is seen that the order of magnitudes agree and there is a good positive correlation (R2 ~ 0.53). To see how well the Ookla speedtest download results agreed with those of the ZDnet's Australia Speedtest (though the name Speedtest appears in both they are not related) we also plotted the two speedtests against one another. As a comparison we also show the ZDnet Speedtest results vs. the PingER NDT.
...