Istanbul 2004
eam member
arger. Team-
‘the task and
figure 2(c),
e agents still
ide only one
rent “Agent
member can
3) The result
oup, and all
In one group
and no team
nically. (5)
oup can take
gent
t will recruit
to complete
ration under
Collaboration
Create
Reuse
ce Agent
ty. Different
operating to
p-level. The
nts or reuse
agents who
ientially or
erver is the
| subtasks to
to up-level
e reused by
;ENTS
nit is agent,
oncurrently.
and parallel
IS function
of interface
cooperative
| exclusive
; are of the
. of a same
et the final
situations:
the other is
resource.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
(II) Heterogeneous cooperative parallel: it is similar to
isomorphic cooperative parallel, but parallel agents are of
different types.
(III) Exclusive parallel: parallel agents are of same type, and
complete a same task. But data resource each agent processes is
different. Some agents may success, and some will fail. Of
course, none knows aforehand which one will success or not.
Only one successful result is needed. Once an agent returns
successful result, all other agents must stop forcibly. If all
agents fail, the task will fail too.
Although parallel is a intrinsic feature of multi-agent system,
how to design parallel algorithms and construct suitable GIS
function agents for real GIS problems in Geo-Agents still must
be paid more attention.
Spatial data is often involved a large spatial scope, and the
content is diversity. So people store spatial data in different
places according spatial scope and content. Distribution is an
intrinsic feature of spatial data. But in another hand, massive
related spatial data for a domain is always stored in a same
spatial database, so spatial data has another feature of
centralization.
The distribution of spatial data brings many inconveniences
because a spatial task always uses many kinds of spatial data
from different spatial databases. But in another hand, just
because of the distribution, many GIS function agents can
cooperate to complete a same spatial task concurrently in
different hosts. This strategy can make the best of distributed
computing resources, and computing can be performed in the
host where spatial data locates, so as to reduce the transferring
quantity of spatial data in network.
Because of the centralization, the needed data resource for a
problem may be in one host. A "Data Priority" strategy (an
agent will try to execute in the host where data resource locates)
is adopted in Geo-Agents, so although a GIS problems can
parallel, the parallel agents will execute place in a same host.
Because the computing resources are limited in one host and
agents also exhaust some computing resources, this parallel will
exhaust more computing resources than sequential execution by
one agent. But from another viewpoint, although agent parallel
in one host couldn't improve the performance, it provides a
simplified and clear structure for constructing applications. For
example, if there are two spatial metadata databases in one host,
it is more convenient and clear to build two spatial metadata
access agents to access different metadata databases
respectively.
According to traditional viewpoint, parallel always improves
performance. In fact, parallel is a kind of collaboration, but
improving performance is not an intrinsic feature of
collaboration. Firstly, collaboration brings new thought for
software construction. Collaboration enables software
construction organized as human society, so complex software
construction can be built more easily and has a clearer
architecture. Secondly, collaboration improves the capacity of
software systems. For a software system, the capacity is
primary. Collaboration makes some tasks achicvable, which
cannot be completed in traditional system. Only after a task can
be complete correctly, improving performance is valid. Of
course, aiming at the disadvantage brought by centralization of
spatial data, a “peer hosts” mechanism is designed to partially
385
improve the performance of Geo-Agents (Luo Yingwei, ef al,
2002).
4. CONCLUSIONS
Collaboration and parallel processing on massive spatial
information in network environment is a key problem that
distributed GIS must face. Agent technology provides a new
effective though and method for processing massive spatial
information in network environment. Geo-Agents is an agent-
based distributed GIS. The collaboration and parallel
mechanisms in Geo-Agents are mainly designed according to
the features of GIS and GIS applications. Through analyzing
two collaboration and parallel samples in Geo-Agents, we can
conclude that Geo-Agents can complete GIS tasks very well,
improve the capability and performance of distributed GIS, and
simplify the development of large complex GIS applications
(Luo Yingwei, 1999).
There are too many collaboration and parallel factors in
distributed GIS. Aiming at the real problems in GIS
applications, how to design collaboration and parallel
algorithms for massive spatial information processing and
implement them in Geo-Agents will be emphases of our future
work.
5. ACKNOWLEDGEMENT
Supported by the National Grand Fundamental Research 973
Program of China under Grant No.2002CB312000; the National
Research Foundation for the Doctoral Program of Higher
Education of China under Grant No. 20020001015; the National
Science Foundation of China under Grant No.60203002; the
National High Technology Development 863 Program under
Grant No. 2002AA135330 and No. 2002AA134030; the Beijing
Science Foundation under Grant No.4012007.
6. REFERENCES
Luo Yingwei, 1999. A Study on Agent-based Distributed GIS.
PhD Dissertation (im Chinese), Beijing: Peking University.
Luo Yingwei, ef al, 2002. The Model of Distributed GIS-
oriented Multi-agent System (in Chinese). Acta Scientiarum
Naturalium Universitatis Pekinensis, 38 (3), pp.375-383.
Hyacinth S. Nwana, 1996. Software Agent An Overview.
Knowledge Engineering Review, | 1(3), pp.205-244.
Ding Xiaoming and Liu Bogin, 1999. Cooperation Mechanism
in MAS. Computer Science (in Chinese), 26(2), pp.54-56.
|
a S
