Massively Distributed Systems
As the world's computing systems become more numerous and more
highly connected, new behaviors appear. As people delegate more
autonomy to programs, and put more trust in their results,
new concerns arise. The Massively Distributed Systems group at
IBM's Thomas J. Watson Research Center conducts research into
some of the implications of a highly-connected world.
Autonomic Computing
We are currently conducting research into the architecture and design
of autonomic computing systems,
working to achieve the vision of
'a network of organized, "smart" computing components that give us what we need, when we
need it, without a conscious mental or even physical effort.'
Computer viruses, epidemiology, computer immune systems
We are the research and development group for IBM's
antivirus
efforts, most recently producing
next-generation virus-protection technologies in association with
Symantec.
We track virus incident statistics,
develop models of worldwide virus spread,
and develop
methods to automatically detect, analyze, and counteract
new viruses and related threats.
See, for instance,
J. O. Kephart, S. R. White, and D. M. Chess, "Epidemiology of
Computer Viruses", IEEE Spectrum, March 1993, Cover and pp 20-26
and
other research papers.
Drawing on analogies with biological immune systems, we have developed an
immune
system for computers and computer networks,
which allows newer, faster-spreading threats to be dealt
with automatically; see, for instance,
J. Kephart,
"A Biologically Inspired Immune System for Computers",
in Artificial Life IV: Proceedings of the Fourth International Workshop
on the Synthesis and Simulation of Living Systems,
edited by Rodney A. Brooks and Pattie Maes, MIT Press, Cambridge, Mass., 1994.
pp. 130-139.
.
Intelligent agents, mobile agents
In agent-based systems, humans delegate some of their
decision-making processes to programs which are (in some
sense) intelligent, mobile, or both.
(See, for instance,
Tim Finin's Intelligent Software Agents page at UMBC.)
"Intelligent" agents have reasoning capabilities, e.g., rule-based
inferencing, probabilistic decision analysis, and/or learning.
See, for instance, our research project on
Embedding Intelligent Agents into Networked
Applications, especially on reusable componentry for
rule-based inferencing and for user authoring and sharing of rule sets,
to perform filtering and re-dissemination of information items such as mail,
news, customer service inquiries, and Web pages.
"Mobile" agents move between different machines to execute their code;
the vision is that this will be
particularly useful for mobile users and mobile communications.
(See, for instance (in postscript)
Colin Harrison's "Smart Networks and Intelligent Agents".)
Are mobile agents a good idea at all?
For some thoughts on the subject, see a recent Research Report
(in postscript),
Are Mobile Agents a Good Idea?.
Another recent report (in postscript),
Itinerant
Agents for Mobile Computing,
also (briefly) addresses some of the security issues involved.
Security in agent-based systems
These agent-based systems also require new
thinking, to avoid both security holes and unexpected
global effects.
See
this overview paper given at a recent Virus Bulletin conference,
this recent book to which we contributed,
and the slightly whimsical
"Things that go bump in the net" page.
When agent-based systems are
combined with
electronic commerce,
the need for all aspects of
security is particularly strong.
Information economies
Today, we are witnessing the first steps in the evolution of the Internet
towards an open, free-market information economy of automated agents buying and
selling a rich variety of information goods and services. We envision the
Internet some years hence as a seething milieu in which billions of
economically-motivated agents find and process information and
disseminate it to humans and, increasingly, to other agents. Over time,
agents will progress naturally from being mere facilitators of
electronic commerce transactions to being financial decision makers
in their own right. Ultimately, inter-agent economic transactions may
become an inseparable and perhaps dominant portion of the world economy.
The goal of the
information economies project
is to anticipate the
likely behaviors of large-scale information economies, and to exploit
this understanding to formulate desirable principles for agents and
agent markets.
Emergent phenomena in distributed systems
When large numbers of programs interact in a connected environment,
various phenomena occur which are not
explicable in terms of the programming or behavior of any
single agent (see for instance
Kephart, J. O., Hogg, T., and Huberman, B. A.,
Dynamics of computational ecosystems, Phys. Rev., 40A, 1989, 404-421.).
It is necessary to understand these phenomena in order
to keep the overall systems both secure and efficient.
(Xerox PARC also has an interesting
dynamics of computation page on the subject.)
Machine learning and neural networks
We are also privileged to provide a home to TD-Gammon, the
best computer backgammon player, and one of the best players
of any species, in the world. See, for instance,
G. Tesauro, "Temporal Difference Learning
and TD-Gammon." Communications of the ACM, Volume 38, no. 3,
pp. 58-68 (March 1995) and
G. Tesauro, "Practical Issues in Temporal Difference Learning,"
Machine Learning vol. 8, pp. 257--277, 1992.
The neural-network techniques learned from experience from TD-Gammon
will be useful in a host of other applications.
David Chess,
chess@us.ibm.com
