Tomography measurement in the Internet

11:31 2008.08.12.

Description of the measurement

topology The goal of the measurement was to investigate a larger tree structure of a real-life network, in which the internal time delays can be resolved from the end-to-end packet-pair travel time distributions. In the measurement the Eotvos University (Budapest) node was the sender. Three pairs of gps synchronized destination nodes were chosen (see the table for the details), each of which 10000 back-to-back packet-pairs were sent to in a round-robin fashion. Between the measurement packet-pairss 0.01 s delays were kept to avoid buffer overflow caused by the measurement stream. The sending of the probes were accomplished by the network interface, the capture was done by DAG cards. A trigger packet was sent to the DAG card of the sender machine to be able to determine the precise time of sending the packet pairs.

-
Nodes we had access to in the measurement
#DestinationPlaceHopsEdges
 Eotvos UniversityBudapest0source
1Szt Istvan UniversityBudapest3+42
1EricssonStockholm3+6+133
2SicsStockholm3+6+93
2TeliaSorenaStockholm3+6+103
3Univerity NavarraPamplona3+6+1+94
3Hebrew UniversityJerisalem3+6+1+44
Legend for the table:
  • # - The id of the packet pair.
  • Hops - How many hops the destination is from the source node. The resolvable inter router hops are also indicated.
  • Edges - How many routes rosolvable are there.

Result and conclusion

timeseries After data collection the end-to-end delay between the sending node and the receivers were determined. (To have a better resolution of the image click on the figure.)

The tree structure of the network was decomposed to have 3 peaces of Y-topologies, each dressed up with two series of one-way-delays from the packet-pair data. Using the tomography technique the distribution of the delay fluctuation on each edges were determined. Then for the remainder edges data could be mined using deconvolution.

results After the analysis for all the 9 edges we had the delay distributions. (For a larger view click on the image.) We may conclude that between the routers where the number of hops are the fewer, the distribution the narrower.




© 2017 The ETOMIC project