The International Computer Science Institute
is pleased to present a talk:

"Practical and Efficient Construction of Network Caricatures"

Prof. Azer Bestavros Boston University
best cs.bu.edu

Friday, October 31, 2003 ICSI, Rm 607 2:30-3:30

The deployment of distributed network-aware applications over the Internet (including the many scientific computing applications over the GRID) requires an accurate representation of the conditions of underlying network resources. To be practical, this representation must be possible at multiple granularities relative to a metric of interest. In the first part of this talk, I will overview MINT, a framework for the efficient construction of such representations using only end-to-end measurements. I will describe a number of instantiations of MINT for various metrics, including packet loss rate, delay, hop-count, and bandwidth. For each such metric, MINT constructions require the availability of a procedure for the measurement of the value of the metric on the shared portion of the paths between a single source and two destinations. In the second part of this talk, I will focus on the development of such a procedure for the bottleneck bandwidth metric.Using novel "Cartouche Probing" primitives, I will show how to efficiently measure the bottleneck bandwidth along *any* segment of a path between two end-points, and how to use such measurements to characterize the bottleneck bandwidth of segments between a set of end hosts. Throughout my talk, I will present experimental results based on simulations and Internet validations, showing the effectiveness of our proposed techniques for building network caricatures.

This work was conducted in collaboration with K. Harfoush and J. Byers.

Speaker Bio:

Azer Bestavros (PhD'92, Harvard U) is Associate Professor and Chairman of the Computer Science Department at Boston University, which he joined in 1991.

Prof. Bestavros' research interests are in the general areas of networking and real-time systems. Prof. Bestavros' networking research aims at improving the scalability of Web and Internet services based on measurement, analysis, and careful redesign. These goals have been fleshed out in a series of projects that culminated in his pioneering of the content distribution model adopted years later by CDNs, his work on Web traffic self similarity and reference locality characterization, his work on various caching and streaming media delivery protocols, and his work on inference of network caricatures using end-to-end measurement. Prof. Bestavros' real-time systems research revolves around improving service predictability and QoS through judicious resource management. To that end, he advanced a number of key concepts, including his generalization of the classical rate-monotonic analysis to accommodate probabilistic guarantees in the presence of uncertainties in resource availability/usage, and his advancement of the use of redundancy-injecting codes to improve the timeliness of client access to periodic broadcasts.