African Internet Topology and Traffic Report
by
Nemo Semret
[email protected]
We present a set of maps showing the state of wide-area Internet
connectivity in Africa. The maps show the routes taken by traffic
from Columbia University in New York to each African node. The time
delay on each "hop" is the average of measurements taken at
6 hour intervals beginning at 22:30 GMT, on Aug 3rd
1998.
While there exist some good connectivity maps of
Africa, we believe we have here the first maps showing the actual
topology, routes, and time delays.
Presented at ESS Telecom 98.
Maps
Observations
The following very preliminary observations can be made.
- On one hand, almost all countries have at least one node with full (TCP/IP) connectivity.
- On the other, outside of southern Africa, intra-regional connectivity is non-existent. The "typical" country has a handful of local nodes, connected to a gateway linked via satellite or cable directly to a major international carrier's network in Europe or North America. Thus all Internet traffic between Ethiopia and Kenya, for example, goes through the US. Given the substantial delay on the long distance lines (typically 500-1000 ms), this means that, even if traffic is very low, the round-trip time (RTT) between neighboring countries is typically more than one or two seconds, i.e. 10 or 20 times slower than the RTT between Europe and North America.
- Africa�s connectivity to the outside is dominated by 5 Internet Service Providers: Alter.net, MCI, Sprint (sprintlink.net and gip.net), the verio group and Teleglobe.
It should be noted that, since these measurements are taken from a single source in New York, some local interconnections may have been missed. However these would have to be very low-speed. Indeed, if there was a direct link from Ethiopia to Kenya, for example, then one expects that at least some of the time, traffic from New York to Addis Abeba would pass through Kenya, whenever the direct route is more congested. By taking a set of measurements that covered every period of the day for 8 days, we expect that very few links were left undiscovered, and any such links would have to be of very low speed or not used for normal traffic. Confirming this hypothesis requires further work, namely tracing routes between African nodes explicitly.
Need for intra-regional connectivity
Shared high capacity link is better than parallel slow links
The most common mathematical model of a data communications link is
the M/M/1 queue. In it, the average delay is given by T =
1/(C-r), where C is the capacity of the link (bits per
second) and r is the average traffic arrival rate (bits per second).
Suppose two similar neighbouring countries each have a long distance
link with capacity C, and have traffic r. They will both experience
the same average delay T. Now if they share a single link with
capacity 2C, and send all their traffic 2r on it, they will both
experience a delay of 1/(2C-2r) = T/2, i.e. half the
delay they had when they both used their own link. This is called
multiplexing gain, and it comes from the fact that,
statistically, they won't be sending data packets at exactly the same
times, so there will be times when one is not sending, and the other
may send and get the full high-speed link alone.
Now since the cost of capacity 2C is generally less than double the
cost of C (most of the cost of a link is independent of the capacity),
it means that each provider spends less money, and gets double the
performance!
Of course, reliability is a good reason for having multiple links.
But still, it is clear that there is ample room for multiplexing gain
in Africa.
Develop intra-regional traffic
For content, commerce, etc., to flourish on the Internet in Africa,
there must be intra-regional traffic. Otherwise, Africans will be pure
consumers and not producers in the information economy.
Acknowledgements, etc.