SAIL

SAIL, short for Statistically Accurate Internet Loss Measurements, is a method for rigorous statistical analysis of packet-loss measurements. It uses algorithms from Hidden Semi-Markov Models to estimate the parameters of the underlying loss process from measurement traces, then computes the loss rate and its variance.

The method is designed to be light-weight: the main algorithms run in linear time in the number of measurement samples. This page collects the SAIL MATLAB implementation, usage notes, sample-data information, and related papers.

Code

The complete MATLAB package is available here:

The archive contains the SAIL library routines together with the main scripts used in the original experiments:

• runSim.m: generates simulated data with a Gamma loss process, then runs SAIL on the simulated traces.
• runExp.m: runs SAIL on experiment data.
• LIBRARY: supporting subroutines for the SAIL implementation.
• TESTSUIT: tests used on the SAIL code.

The README describes the expected data layout for runExp.m: for each measurement there should be two MATLAB files, one containing binary probe outcomes and one containing the corresponding probe sending times.

Measurement Data

The original SAIL experiments used more than 50 GB of loss-measurement data. The old project page made small text-format samples available for DNS, Web, and UA-EPFL measurements, and noted that the full dataset was available on request.

Because some of the sample archives are large enough to be awkward for normal GitHub and Netlify deployment, they are not copied directly into this repository. For access to the complete dataset, please contact the SAIL authors.

Papers and Presentations

• Rigorous statistical analysis of Internet loss measurements, H. X. Nguyen and M. Roughan, IEEE/ACM Transactions on Networking, 21(3), 734-745, 2013.
• Statistically Accurate Network Measurements, Isaac Newton Institute presentation, 2010.
• Statistically Accurate Internet Loss Measurement, Hung X. Nguyen and Matthew Roughan, Technical Report, The University of Adelaide, Australia.
• Statistically accurate Internet loss measurements, H. X. Nguyen and M. Roughan, in Proceedings of ACM SIGMETRICS 2010, short paper, New York, USA.
• SAIL presentation.
• On the Correlation of Internet Packet Losses, H. X. Nguyen and M. Roughan, in Australasian Telecommunications Networks and Applications Conference, pp. 22-27, Adelaide, Australia, 7-10 December 2008.

Acknowledgements

The original MATLAB implementation uses two existing scripts:

• The HSMM forward/backward implementation hsmm_new.m, by Shun-Zheng Yu.
• The inverse Laplace transform implementation invlap.m, by Karl Hollenbeck, Department of Hydrodynamics and Water Resources, Technical University of Denmark.