prediction tools for tstat data
Branch: master
Clone or download
Latest commit 6b53853 Jan 7, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
prediction+IPs new files Dec 20, 2018
prediction+smoothing new files Dec 20, 2018
license.txt Create license.txt Jan 8, 2019

Research Network Transfer Performance Predictor (netperf-predict)

This respository containts two sets of analysis routines for predicting the percentage of retransmitted packets on network flows. One directory contains code that applies random forest regression in order to predict the number of retransmitted packets on each flow, operating on timeseries data from the tstat tool, which outputs flow-like data. The second directory also applies a random forest regression and also incorporates a "smoothing" routine that increases accuracy in some situations.

This code was developed at Lawrence Berkeley National Laboratory as part of the NSF IRNC-funded "NetSage" project, award number OAC-1540933 (Lead PI at Indiana University, Jennifer Schopf; Co-PI and Berkeley Lab Lead, Sean Peisert).

Results of using this code are described in the following paper:

Anna Giannakou, Daniel Gunter, and Sean Peisert, "Flowzilla: A Methodology for Detecting Data Transfer Anomalies in 
Research Networks," Proceedings of the 5th Innovate the Network for Data-Intensive Science (INDIS) Workshop, Dallas, TX, 
November 11, 2018.

Download PDF

Contributors to this code repository at the Berkeley Lab included:

Questions about the project to the code's contributors are welcome.