You are here



DMFSGD: a tool to infer end-to-end network performance from a few available measurements in a fully decentralized manner. DMFSGD is developed at Research Unit in Networking of University of Liège. Mathematically, DMFSGD formulates the network performance prediction problem as a matrix completion problem and solves it by matrix factorization. It provides a generic and flexible framework to deal with various performance metrics, such as Round-Trip Time (RTT), Available Bandwidth (ABW), expressed either as values or discrete classes (i.e, ratings). The Figure shows the architecture of DMFSGD for the prediction of network performance classes.

Architecture of DMFSGD for predicting network performance classes.


For more details, we recommend the users to refer to our papers:

Wei Du, Yongjun Liao, Narisu Tao, Pierre Geurts, Xiaoming Fu and Guy Leduc, Rating Network Paths for Locality-Aware Overlay Construction and Routing, IEEE/ACM Transactions on Networking, to apear 2015.

Yongjun Liao, Wei Du, Pierre Geurts and Guy Leduc, DMFSGD: A Decentralized Matrix Factorization Algorithm for Network Distance Prediction, IEEE/ACM Transactions on Networking 21(5), 2013.

Yongjun Liao, Wei Du, Pierre Geurts and Guy Leduc, Decentralized Prediction of End-to-End Network Performance Classes, The 7th International Conference on emerging Networking EXperiments and Technologies (ACM CoNEXT), 2011, Tokyo Japan.


Quick start:

The Matlab implementation of DMFSGD is available at: A detailed Readme file is included that explains how to run DMFSGD.


New features supported by the mPlane project

Thanks to the support of the mPlane project, we implemented DMFSGD for RTT prediction in Python and tested it on the platform of Planetlab.


mPlane proxy interface

DMFSGD has no mPlane proxy interface.


Official version
  • May 15th, 2014: ADD TARBALL frozen release for D2.2