We develop a new model for trading in large-scale distributed computin
g environments which is based on the gradual evolution of a federated
trading space through a process of continual exploration and evaluatio
n, rather than on the imposition of a strictly managed structure. In o
ur model, each trader autonomously acquires local knowledge of the tra
ding space, called trading knowledge, through a process of distributed
resource discovery. Trading knowledge typically consists of trader li
nks which reference other traders. Trader links also contain a measure
of affinity: a strength of attraction based on a comparison of so-cal
led service and interest profiles, perhaps combined with a history of
how useful and reliable other traders have proved to be. This notion o
f affinity helps a trader to decide how to resolve import requests whi
ch cannot be satisfied locally. It also helps a trader to decide which
trader links to retain as it periodically and autonomously explores t
he trading environment. Instead of being concerned with the detailed m
anagement of individual trader links, human managers can then control
the evolution of the trading space through a number of high-level mana
gement policies. These include service and interest profiles, definiti
ons of affinity and instructions on when and how exploration should oc
cur. Our paper also describes a reference implementation of the model
with the ANSAware distributed processing environment called the Explor
ative Trading Service (ETS) which provides an exploration engine as we
ll as various management interfaces (including a Gopher gateway so tha
t various network information retrieval browsers may be used to browse
the trading space). (C) 1998 Published by Elsevier Science B.V.