LU operator measurement reports (from serveurDEDIE)

Choosing an internet access provider (either permanent or dialup) is a very difficult task. Giving advice for this task is equally difficult. We have faced this problem when our customers asked which was the best provider to reach our network. Some also wanted to know how visible their web servers would be from the major providers in their country. To answer these questions we started did to do a trivial thing: document our own connectivity. Then we added a traceroute gateway and setup a measurement system to provide data on the major IAPs.

This tool enables direct comparison of the roundtrip and the packet loss to operators of the measurement that occured in the last 15 minutes. It also features a history of the last few hours. The system computes the data for each operator every 15 minutes. This is done with the ping function, with 15 packets of 568 bytes. Pings are sent to a DNS server in each sampled network. DNS servers were choosen as sampled hosts because they are not routers and most do not discriminate ICMP packets. DNS servers are also strictly monitored for availability. On request we occasionaly ping boxes that are not DNS servers.

This ping tool generates accessibility data of the sampled host from our server, but comparison on a single set of measures does not makes sense. We therefore decided to archive the data and prepare monthly reports. In these reports we focus on the successful packet transmission rate (opposite of packet loss rate) and the reachability rate (opposite of unreachable rate). We believe those are the two major indicators of quality. Packet loss is due to congestion, i.e. when a link gets overloaded with packets it has to discard some. Unreachable situations are caused when a router does not have any route left to a destination (either BGP or IGP). This is caused by unstable routing often due to physical link loss (fiber cut), latency of routing protocols (routing loop), human configuration errors, etc.

We also have roundtrip measurements, but we consider the results as meaningless.

Interpretation

The validity of our data is subject to numerous contesting, we are aware. Doing internet measurement comparisons is a hazardous exercise. There's no proper way of doing them. "Ping" measurements have their detractors. Of course, our measurements are influenced by our own connectivity and our BGP configuration and our local network (10/100 etherswitched cisco lan). Nevertheless our multiple connectivity and our location in Washington maximise the fairness of the measurements (because we are on the East coast, close to transatlantic links, with multiple links not to discriminate between the different major backbones). And of course, our statistics include possible availability problems with the DNS servers.

Our measurements are not end to end. That's also a feature. Measuring the part between the backbone and the leaves of an operator is extremely complex because of the multiple media involved (PSTN, ISDN, Leased Line, ADSL).

In our reports we only evaluate the reliability and performance between our network and sampled hosts of european operators. And of course: those results do not mean an operator is better than another. For any question related to these measurements, you can join the connectivity mailing list or write to rapports@TBS-internet.com

Results


[ TBS-internet ] [ HOME ] [ statistics ] [ connectivity audit ]

© TBS-internet, 2000, all rights reserved. No reproduction, copy or mirroring allowed.
Revision date: 19 apr 2000
URL: http://www.serveurDEDIE.com/stats/LU/rapports/intro.html