| GENERAL
BACKGROUND
Amorphous computing brings together large
numbers of processors or processes into an irregular configuration that
is non-prescribed, dynamic, and possibly unknown or even unknowable, in
order to produce an entity capable of performing some set of tasks. In
addition to efficiency, performance criteria include adaptability/evolvability,fault
tolerance/error correction, and robustness. Such systems are now in
their infancy, but as processor size/cost falls and networks grow, our
ability to effectively organise them without recourse to centralised
control will become a key challenge. Biological systems scan, process,
store and transmit information via interactions between a large number
of units performing simple tasks as part of highly variable networks.
Such systems are important exemplars of amorphous computers existing at
a range of scales and involving units of variable complexity. We will
study three key examples across this range: neural, epidemic and
proteomic networks. Through the use of mathematical and computational
models, we will address their structure, dynamics, development, and
adaptation, deriving principles and paradigms for the design and control
of amorphous computers.
|
Our biologically inspired results will be relevant to a number of
outstanding issues in computer science, where scalability problems, on
the one hand, and unpredictable and/or changing conditions, on the
other, dictate that explicit global configuration is either impossible
to specify or obsolete by the time it is implemented. Well documented
application domains range from ad-hoc sensor arrays comprising very
large numbers of simple devices, resource allocation in Grid computing30
and
management of machine-to-machine interactions in autonomic51
systems,
to the design of workable peer-to-peer (P2P) architectures.13,22,56
Such
problems naturally lend themselves to an amorphous computing approach,
where the challenge can be formulated in terms of an ability to
automatically and autonomically reorganize reorganize overlay networks
(i.e., impermanent, potentially ad-hoc networks of functional
relationships superimposed on and instantiated over a typically more
permanent preexisting network of connectivity). More
background information can be found in
backgroundinfo.pdf
|
|
