Abstracts
Anthony Finkelstein
This seminar reviews the technical and decision-making processes by
which the Cambridge Financial System came into being. The talk
outlines the history of the CAPSA project and discusses key issues
arising from the system development. The application of software
engineering practice in the implementation of large Enterprise
Resource Planning (ERP) systems is discussed and some research issues
signalled.
Richard Durbin
Tim Harris
Through careful design and implementation it's possible to build data
structures that are safe for concurrent use without needing to manage
locks or block threads. These "lock-free" data structures can
increase performance by allowing extra concurrency and can improve
robustness by avoiding problems such as priority inversion. However,
many existing designs are extremely intricate and make unrealistic
assumptions that render them effectively unusable.
In this talk I'll introduce the topic and present some of the work
that we've been doing on practical designs for lock-free data
structures, on their implementation and on their evaluation within
real systems. The subject is one of broad appeal, encompassing both
checking the algorithms' correctness and handling the implementation
issues that emerge on modern processors with relaxed memory ordering.
Simone Teufel
I will discuss my work on the summarisation of scientific articles
which is based on a robust model of the discourse structure of this
text type. Training from surface linguistic and non-linguistic clues,
it is possible to identify sentences of a certain rhetorical status --
goal statements, criticism of previous work or statements of research
predecessors. These pieces of information help to situate one given
article in the scientific field, and to distinguish it from similar
articles, in a far more informative way than a large citation index
could do.
One of the challenges for the future lies in the production of
coherent, flexible summaries from the extracted sentential
material. One interesting starting point for this task will be the
relationships between sentences from the same article which are of the
same rhetorical status.
Part 1B Group Project Presentations
Richard Gold
We examine situations where the typical IP network layer model lacks
certain functionality that certain applications desire: namely content
location and support for different routing styles. We describe the
approaches that we have taken to overcome these limitations. We
overlay IP to provide content location and undermine IP to allow the
expression of alternative routing paradigms. Overlaying entails the
construction of additional routing structures on top of the existing
IP infrastructure. This allows us to create an overlay network which
consists of a subset of the IP network nodes. The result of this is
that it allows us to take control of the routing infrastructure and
thus make decisions which IP is unable to take. A natural consequence
of such an approach is better network support for overlay networks. We
propose the usage of an underlay network (i.e., a virtual network
below IP) for such tasks. In joint work with Christian Tschudin from
Uppsala University, we present the SelNet underlay network and
describe its ability for allowing the creation of overlay networks and
we propose the basic component of the overlay network: the "tunnel" as
a core networking abstraction and show how it can be used to create
various styles of routing.