Stanford Networking Seminar

12:15PM, Thursday April 7, 2011
Gates 104


High-Fidelity Latency Measurements in Low-Latency Networks
 

Ramana Kompella
Purdue University


About the talk:
For the longest time, networking engineers and researchers have focused mainly on obtaining high end-to-end throughput in IP networks. In recent years, however, latency has evolved into a metric that is as important as throughput in IP networks. While low latency is a desirable property for any network-based application, this obsession towards low end-to-end latency stems from the stringent requirements of many new kinds of datacenter, cloud, and wide-area applications that have become popular in the recent times. Network operators managing such latency-sensitive applications need sophisticated tools for high-fidelity latency measurements at various places in the network that will help them identify root causes of SLA violations, determine offending applications that may hurt the performance of others, perform traffic engineering and so on. Unfortunately, routers offer no native support for tracking latencies even within the router making it hard for network operators to measure and pin-point latency spikes in the network. In this talk, I will describe our recent work in devising scalable router primitives for fine-grained high-fidelity latency measurements. Specifically, the talk will focus on two major techniques: (1) Reference Latency Interpolation (RLI) architecture for obtaining latency measurements on a per-flow basis; and, (2) MAPLE, an architecture for storing latency of each and every packet in a scalable fashion and computing latency statistics across arbitrary subpopulations.

About the speaker:
Prof. Kompella's main research area is computer networks. Some recent topics of interest include scalable router primitives for measurement, fault isolation in backbone and enterprise networks, and improving network performance in virtualized and cloud environments. Dr. Kompella joined Purdue in Fall of 2007 as an Assistant Professor in Computer Science after obtaining his Ph.D. degree from UCSD. He obtained his Masters degree from Stanford University in 2001 and an B.Tech degree from IIT Bombay in 1999, both in Computer Science. He is a recipient of the NSF CAREER Award in 2011.