up previous


Bhaskaran Raman 
 
475 Soda Hall, Apartment G,
EECS Dept / CS Division, 1836 Hearst Avenue,
U.C.Berkeley, CA 94720. Berkeley, CA 94703.
Phone: +1-510-642-8284 Phone: +1-510-644-1906
Fax: +1-510-643-7352
bhaskar@cs.berkeley.edu
http://www.cs.berkeley.edu/~bhaskar/
Research Interests Computer networks, large-scale Internet-based systems, Internet middleware for heterogeneous network integration; cross-cutting issues in networking and distributed systems; mobile and wireless networks. I'm also particularly interested in the development of appropriate technology.
Teaching Interests Graduate as well as undergraduate teaching; computer networks and protocols, Internet-based distributed systems, wireless networks, operating systems.
Education Research Experience Teaching and Mentoring Experience Publications Theses Selected Posters and Presentations Professional Activities
Student member of IEEE (Communications Society, Computer Society),
Student member of ACM
Refereed papers for ASPLOS 1998, ACM Mobicom 1998/1999/2000, WMCSA 2000, IEEE Infocom 2000/2001, IEEE Personal Communications Magazine 2001.
Honors Other Activities
I'm an active volunteer for ``Asha for Education'' (http://www.ashanet.org/), an organization that works to promote basic education in India.
References PhD Thesis Summary
An Architecture for Performance and Availability Constrained Service Composition in the Wide-Area Internet
Service Composition offers flexible ways to quickly enable new application functionality by putting together existing components. Unix pipelining is a very simple example of composition. In my thesis, I focused on the scenario where there are multiple service providers that develop, implement, deploy, and manage different component services at different points on the Internet. Other portal providers compose a set of these services to enable new applications. An example of such a model in today's Internet is the Yahoo portal making use of a third-party search engine to provide an end service to its subscribers. Such a model is likely to emerge given the importance of services in 3G+ networks. There are several multimedia applications that can built on this framework of composition - for instance, we built an application that was capable of reading out news to a cell-phone user, by composing a news source service with a text-to-speech conversion engine component. While past work has addressed performance and fault-tolerance issues in service composition within a single service provider cluster, composition in our scenario brings up new challenges. Since service providers are independent, the components could be spread across the wide-area Internet. This makes it challenging to choose a set of service instances for a particular client session (more challenging than the problem of single web-mirror selection), based on current network and server performance. Further, studies by Labovitz et.al. have shown that Internet path availability can be very poor, and also inter-domain path recovery can take several minutes. Improving this availability by detecting and recovering from failures quickly is a challenge that I address in my work. In my thesis, I have proposed and prototyped an architecture to address these performance and availability issues in service composition. My architecture is based on an overlay network of service execution platforms. The execution platforms form the middleware layer on which providers deploy their services. And the overlay network provides the context for exchange of network/server performance information exchange - this enables an "optimal" choice of service instances for client sessions. Further, the overlay network also allows detection of network path failures, and the subsequent choice of alternate service instances. Some of the key design features of this architecture are: (a) the separation of service location information from that of performance/liveness information, (b) the client session is setup as a virtual-circuit in the overlay network - this means that session recovery on failure need not depend on the propagation and stabilization of the failure information across the network, and (c) the use of soft-state to maintain the virtual circuit - this makes it easy to implement session tear-down in a distributed fashion. I have built a prototype of the system, and evaluated it using an emulation platform - there is an actual implementation of the algorithms and interfaces, and the wide-area network characteristics are emulated to study issues of time-to-session-recovery, scaling, and stability. My evaluations show that multi-media sessions can be recovered within a few seconds after wide-area network failure. This represents a significant improvement over Internet-path recovery which takes several 10s of seconds to several minutes.

About this document ...

This document was generated using the LaTeX2HTML translator Version 2K.1beta (1.47)

Copyright © 1993, 1994, 1995, 1996, Nikos Drakos, Computer Based Learning Unit, University of Leeds.
Copyright © 1997, 1998, 1999, Ross Moore, Mathematics Department, Macquarie University, Sydney.

The command line arguments were:
latex2html -split 0 -nosubdir bhaskar-cv.tex

The translation was initiated by Bhaskaran Raman on 2002-11-11


up previous
Bhaskaran Raman 2002-11-11