Network Monitoring with AMP
C2 | Thu 24 Jan | 2:25 p.m.–3:10 p.m.
Brendon is a research programmer at the WAND Network Research Group where (amongst other things) he is the primary developer of the AMP measurement software. He has long standing interests in active network measurement, as well as visualisation, which his work on AMP allows him to combine. He's been involved in every aspect of the measurement process, from constructing the raw packets to be sent out on to the wire, through to storing, processing and displaying the aggregated data on the web for user consumption. Most recently Brendon has been working on writing tests using headless web browsers to measure video playback quality, and investigating ways to make BGP routing software faster (parallelism!) and smarter (with centralised policy and control).
AMP is a distributed network monitoring platform designed to continuously perform active network measurements, capable of monitoring both network and service performance. Currently available tests include ICMP and TCP latency, UDP DNS latency, TCP throughput, HTTP performance and video streaming performance. Scheduled tests run continuously on the probes and report data to a central collector (or into your favourite storage/graphing tool) to generate graphs. The same tests can also be run as standalone tools from the command line as required. There are AMP probes deployed at most universities in New Zealand, at major sites on the REANNZ network, and at many New Zealand Internet Service Providers, all with publicly available data. AMP is also being used internally by a number of ISPs to monitor customer experience and infrastructure performance. This talk will cover the architecture and design of the AMP platform, as well as how the tests work, how the data gets reported and some of the issues we've encountered while trying to take accurate measurements. In particular it will show some of the challenges involved in creating a test to measure video streaming performance using a headless web browser. It will also explore some of the data that has been collected over the last few years and showcase some of the more impactful network events, misconfigurations, or generally interesting behaviour that has been observed.