[Colloq] Hiring Talk *Monday, March 21* Patrick Eugster, SUN Microsystems

[Rachel Kalweit]

College of Computer and Information Science Colloquium

Presents:
Patrick Eugster
SUN Microsystems, Switzerland

Who will speak on:
Abstractions and Algorithms for Pervasive Programming


Date: Monday, March 21, 2005
Time: 10:30am
Location: 366 West Village H
Northeastern University

Abstract:
Through the continuously increasing number of devices equipped with 
Computing and communication capabilities, the vision of "pervasive 
computing" is materializing. Computing has become inherently 
distributed, but with the power of pervasion comes complexity, as 
typical applications involve more and more components, both hard- and 
software ones. This places yet more burden on the application developer, 
which namely has to deal with aspects of the underlying infrastructure 
such as (lacks of) availability and reliability of communication and 
hosts, in addition to software aspects, such as extensibility, 
reusability, and adaptability of software components. These constraints 
ask for new programming paradigms which account for the dynamism of 
modern distributed settings. A fundamental concern consists in 
minimizing coupling of components.

This talk presents abstractions and algorithms for programming in 
pervasive settings, centered around a candidate abstraction called 
type-based publish/subscribe (TPS). TPS combines the loose coupling 
provided by the publish/subscribe communication style with high-level 
guarantees such as type safety and encapsulation. We present an overview 
of two ways of making TPS accessible to the programmer in the Java 
programming language, a first one based on specific primitives added to 
the language, and a second one making use of recent inherent features of 
Java. Being essentially a multicast primitive, TPS is implemented mainly 
through multicast algorithms. We outline a set of such algorithms --- 
designed and implemented in the context of TPS --- which follow a 
randomized approach. Through these algorithms, we discuss the benefits 
of probabilistic approaches for pervasive computing, as well as the 
difficulties encountered when providing probabilistic guarantees to the 
programmer.

Host: Karl Lieberherr