Biomedical Engineering Reference
In-Depth Information
mone, is used to control the activity of a single
agent and better accomplish the system goal.
The ARMAP pick and drop probability func-
tions are inspired by probability functions that
were introduced in (Deneubourg & al., 1990), and
later elaborated and discussed in (Bonabeau &
al., 1999), to emulate the behavior of some spe-
cies of ants that build cemeteries by aggregating
corpses in clusters. In ARMAP, such functions
have been adapted through the following main
modifications: (i) descriptors are not only moved
and aggregated, as are corpses in the mentioned
papers, but also
replicated
; (ii) descriptors are
reorganized according to the class to which they
belong, so as to reduce the overall entropy of the
system.
number of agents is maintained to a value which
is about
Ngen
times the number of hosts.
Periodically each ARMAP agent sets off from
the current host and performs a number of hops
through the P2P links that interconnect the Grid
hosts. Then the agent uses random
pick
and
drop
functions in order to replicate and move descrip-
tors from one peer to another. More specifically,
at each host an agent must decide whether or not
to
pick
the descriptors of a given class, and then
carry them in its successive movements, or to
drop
the descriptors that it has previously picked
from another host.
As a consequence of pick and drop opera-
tions, each host can maintain descriptors of local
resources, which are never removed, as well as
descriptors of resources published by other hosts,
which can be picked up and discarded by agents.
In the following, when distinction is relevant,
such descriptors will respectively be referred to
as
local
and
remote
descriptors.
Pick and drop probability functions are dis-
cussed in the following.
DISSEMINATION OF RESOURCE
DESCRIPTORS
The aim of the ARMAP protocol (Forestiero
& al., 2005) is to achieve a logical organization
information on the Grid by spatially mapping
resource descriptors according to the semantic
classification of related resources. It is assumed
that the resources have been previously classified
into a number of classes
Nc
, according to their
semantics and functionalities (Crespo & Garcia
Molina, 2002).
The ARMAP protocol has been analyzed in
a Grid in which hosts are connected in a P2P
fashion and each host is connected to at most 8
neighbor peers. The Grid has a dynamic nature,
and hosts can disconnect and rejoin the network.
When connecting to the Grid, a host generates
a number of agents given by a discrete Gamma
stochastic function, with average
Ngen
, and sets
the life time of these agents to
PlifeTime
, which
is the average connection time of the host, calcu-
lated on the basis of the host's past activity. This
mechanism allows for controlling the number
of agents that operate on the Grid: indeed, the
Pick Operation
Whenever an ARMAP agent hops to a Grid
host, it must decide, for each resource class,
whether or not to
pick
the descriptors of this
class which are managed by the current host,
unless it already carries some descriptors of the
same class. In order to achieve replication and
mapping functionalities, a
pick
random function
is defined with the intention that the probability
of picking the descriptors of a class decreases as
the local region of the Grid accumulates these
descriptors and vice versa. This assures that as
soon as the equilibrium condition is broken (i.e.,
descriptors belonging to different classes begin
to be accumulated in different regions), a further
reorganization of descriptors is favored.
The basic
Ppick
probability function is shown
in formula (1). The agent evaluates this function,
Search WWH ::
Custom Search