Stanford University Networking Seminar
Historically, packet network traffic diagnostics have relied on 1) counts of packetsbytesdrops/etc. from switches and routers and 2) active probes used for RTT measurements and for goodput and retransmits within the probe flows. Recently, researchers and network operators have started building diagnostics using passive monitoring (via packet header capture) of production traffic. These current efforts extract only a fraction of the wealth of information available and impose severe constraints on how much must be monitored and where (e.g., both directions of a flow must pass through the capture point and all traffic, not just samples, must be inspected) and sometimes how the packet capture is done (e.g. requiring specific switch and port types).
Pollere has developed purely passive techniques and diagnostics that mine critical transport performance measures (RTT, delay, delay variation, throughput, loss) from a packet header stream that may be sampled and/or unidirectional and taken anywhere in the network. This talk will describe that work using an analysis of the detailed traffic behavior of different video providers (YouTube, Netflix, Amazon, Apple) to make the ideas concrete.
Kathleen Nichols is CTO at Pollere, a network consulting and contract research company. Her current interests and work is on extracting performance data from passive packet header capture and she has worked on bufferbloat solutions, differentiated services, and dabbled in information centric networking. She received a PhD in EECS from UC Berkeley a very long time ago.