International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
authorization is required. AgentServer manages a table to
register other “Agent Regions" that are authorized with itself
each other. Between two directly authorized *Agent Regions",
agents between them can communicate via the coordination of
their AgentServers. The authorization relation in Geo-Agents
can be passed one by on: If a directly authorized AgentServer
chain can be found for two “Agent Regions” that have no direct
authorization, agents between the two “Agent Regions” can
communicate via the coordination of the AgentServer chain.
2. COLLABORATION MECHANISM IN GEO-AGENTS
In human society, there are some relations among individuals,
which make individual collection become human society, and
make individuals become socialization people. Just like human
society, multi-agent system must let its agents cooperate with
each other to become an integrated system, so as to complete
more complicated tasks (Hyacinth S. Nwana, 1996; Ding
Xiaoming and Liu Bogin, 1999).
The goal of Geo-Agents is to construct a multi-agent system
like human society, so collaboration mechanism is necessary.
There are two collaboration modes in Geo-Agents:
collaboration among interface agents and collaboration under
control of interface agent.
2.1 Collaboration among Interface Agents
Collaboration among interface agents is coordinated by
AgentServer. AgentServer manages a table to register different
cooperation groups in one “Agent Region”. There are three
types of collaboration among interface agents: direct
cooperation, peer group cooperation and charge-tributary group
cooperation.
(I) Direct cooperation: during the execution of an interface
agent, it needs some results of other interface agents (see figure
2(a)).
(e) Charge-tributarv Group Cooperation
Figure 2 Collaboration among Interface Agents
(IT) Peer group cooperation: several interface agents form a
team. All team members cooperate to complete a same complex
task. Each member assumes a subtask, and accomplishes it
independently. Team members are peer in the team: after
completing subtask, each member sends its result to all other
members, so each member can obtain a same final result of the
task (sce figure 2(b)).
(II) Charge-tributary group cooperation: like peer group
cooperation, all team members cooperate to complete a same
384
complex task. But there is a team-charger. Team member
completes subtask and sends its result to team-charger. Team-
charger assembles all sub-results to final result of the task and
sends the final result to all team members (see figure 2(c),
where A is team-charger).
In Geo-Agents, the collaboration among interface agents still
has some restrictions: (1) Collaboration exists inside only one
"Agent Region". Collaboration among different "Agent
Regions" is not supported now. (2) Each team member can
communicate with others only once in one group. (3) The result
type of each team member is same in one group, and all
members cooperate to complete a same task. (4) In one group
team, the number of team member is pre-fixed, and no team
member can join into the team dynamically. (5)
Communication among interface agents in one group can take
place only after all team members are joined in.
2.2 Collaboration under Control of Interface Agent
After an interface agent accepts a complex task, it will recruit
some GIS function agents, and then organize them to complete
the task cooperatively. This is hierarchy collaboration under
control of interface agent (see figure 3).
| Interface Agent | — Interface Agent — Interface Agent |
AC A
A T A,
oe E — Create
Ay A An| — N me Reuse
——— Collaboration
Figure 3 Collaboration under Control of Interface Agent
The hierarchy of agents is similar to human society. Different
agents play different roles in the system when cooperating to
complete a same task. Interface agent locates at top-level. The
up-level agents may create some under-level agents or reuse
other existing agents to complete subtasks. Those agents who
belong to a same parent can execute sequentially or
concurrently.
In hierarchy collaboration: 1) Facilitator/AgentServer is the
collaboration coordinator; 2) up-level agents assign subtasks to
down-level agents and down-level agents return to up-level
agents; and 3) the results of some agents may be reused by
others.
3. PARALLEL MECHANISM IN GEO-AGENTS
In a multi-agent system, the smallest executing unit is agent,
and many agents can execute independently and concurrently.
Geo-Agents likes a kind of cluster parallel system, and parallel
is performed by GIS function agents. Parallel of GIS function
agents takes place in collaboration under control of interface
agent, and exists as three modes: isomorphic cooperative
parallel, heterogeneous cooperative parallel and exclusive
parallel.
(D Isomorphic cooperative parallel: parallel agents are of the
same type, and each of them completes a subtask of a same
complex task. Assembling all subtask-results will get the final
result. Isomorphic cooperative parallel meets two situations:
one is that all agents process different data resource; the other is
that all agents process different parts of a same data resource.
Interi
(ID
isom
diffei
(IM)
comp
diffei
cours
Only
succe
agent
Althe
how
funct
be pa
Spati
conte
place
intrin
relate
spati:
centr
The
beca
from
becat
coop
diffei
com
host |
quant
Beca
probl
agent
Is ad
paral
Beca
agent
exhai
one a
in or
simpl
exam
it is
Acco
perfo
impr
collal
softw
const
const
archi
softw
prime
canne
be ce
cours
spatie