The paper presents an overview of AEGIS framework. It lists the
framework's features and discusses its architecture from a very high
level of abstraction. It also provides some details on indexing
structures implemented within the framework and shows some results of
experiments performed with the use of this framework.

1. The chosen style (and level of abstraction) allowed the autors to
omit some facts that would be interesting for the reader: a) It is
written that "The data model is based on the Open Geospatial Consortium
(OGC) Simple Feature Access (SFA), OGC Spatial Referencing by
Coordinates and ISO19108 Temporal Schema". Does it mean, that classes
implemented within the scope of the AEGIS framework match 1:1 the
classes described in the mentioned specification? From the description
provided it can be deduced, that this is not the case (this is not a
direct 1:1 mapping). The question was rised because there exist well
known frameworks, that support OGC SFA or can be mapped into it (for
example FDO (http://fdo.osgeo.org/), GeoTools
(http://www.osgeo.org/geotools)). A few sentences including comparison
to these frameworks would clear the situation. b) "Graph representation"
is difficult to uderstand. c) In IT world there is a quite difficult
problem with ORM mapping (mapping from object model used by programming
frameworks into relational model used in databases). The authors
mentioned that they use as a backend MongoDB (which is document
database) and that support for PostGIS (which is object-relational
database) is planned. Thus the question rises - what data model is
implemented on the database side? Short explanation would be welcome. d)
The interesting issue is "how the rasters are represented on the
database side and which interface is used to access them efficiently".
Would it be possible to privide some sentences with explanation? Maybe a
comparison to GRASS storage would help?

2. Image processing algorithms can work "in place" (like thresholding)
or "out place" (like median filtering). This is important for
implementation of efficient management of resources. Is this issue
considered within the framework?

3. The article could be better shaped with the following corrections: a)
The sentence: "Figure 2 shows one layout possibility using thin and
thick clients with centralized data storage." should be moved to the
other place (maybe to the Communication list item on page 4?). b)
Experiment with path planning: A* algorithm is a graph search algorithm,
with backtracking, involving heuristics for nodes selection. Assuming
that temporal information is not changing (travel time for a particular
road segment) an agent can plan the whole route beforehand (before any
move). However, if the temporal information is changing during the time
(because of the traffic) then such "whole route planning" is not
possible. This rises a question - do the agents run in a "whole route
plannig" mode? If yes, there should be a correlation between temporal
information and congestion observed. The road segments with lower
"multiplication parameter" should be choosen by the agents more often
then other segments. And since that the congestion observed contradicts
the "multiplication parameters" assigned to the road segments. This
phenomena can be observed in many multi agent systems (MAS) that share
the same knowledge. Thus, the experiment described in the paper should
be enriched with a few comments higlighting the meaning of the results
obtained. Otherwise the figure 7 can not be interpreted well by the
reader without prior knowledge of A* and MAS. If no, there are several
issues related to the route planning in MAS (with agents that runs in
parallel, interfere and change the temporal data). But these issues are
out of the scope of the paper. c) Eperiments with image segmentation
Figure 8 without description does not provide any useful information
other then "our algotihms work". Thus this figure should be withdrawn
from the paper. d) Spelling: sinlge (page 6) shich (page 14)