Sapan Bhatia
Princeton University
PlanetFlow2 is a system that continuously logs PlanetLab's traffic, periodically transfers it to a central location and allows users, administrators and third parties to query the data. PlanetFlow2 1 was implemented with two primary goals: to support an advanced querying mechanism that would allow traffic logs to be aggregated based on one or more header fields, and to be fast and scalable. A secondary set of goals was to make it robust so that it would function even when high-level components of PlanetLab such as the Node Manager were not functioning as expected. PlanetFlow2 runs continuously on all of PlanetLab, collecting about 1TB of data every month, and accounting for about 120TB of data at any given time. It can respond to complex queries such as 'How many unique ports did Slice X contact?', or 'List the slices that contacted over 1000 unique IP address', but which typically complete within a 5-20 seconds. This document describes the usage of PlanetFlow, its installation and its design.