Geography Reference
In-Depth Information
former. MapProxy offers three different ways to describe the extent of a seeding or cleanup
task: a simple rectangular bounding box, a text file with one or more polygons inWKT format,
or polygons from any data source readable with OGR (e.g. Shapefile, PostGIS).
These three services support both metatiling and meta-buffer methods. The meta-buffer adds
extra space at the edges of the requested area.
When a request of a tile in an unsupported coordinate reference system (CRS) is received, both
GeoWebCache and MapProxy supports the reprojection on the fly from one of the available
CRSs to the specified one. The former achieves this using GeoServer, while the latter offers it
natively.
5. Cache management algorithms
Significant improvements can be achieved by using a cache of map tiles, like the ones
discussed above. However, adequate cache management policies are needed, especially in
local SDIs with lack of resources. In this section, our contributions to the main cache strategies
are presented: cache population (or
seeding
), cache replacement and tile prefetching.
5.1. Cache population
Anticipating the content that users will demand can guide server administrators to know
which tiles to pregenerate and to include in their server-side caches of map tiles. With this
objective in mind, a predictive model that uses variables known to be of interest to Web map
users, such as populated places, major roads, coastlines, and tourist attractions, is presented
in [8].
In contrast, we propose a descriptive model based on the mining of the service's past history
[7]. Past history can be easily extracted, for example, from server logs. The advantage of this
model is that it is able to determine in advance which areas are likely to be requested in the
future based exclusively on past accesses, and it is therefore very simple.
In order to experiment with the proposed model, real-world logs from the IDEE-Base
nation-wide public web map service have been used. Request logs were divided in two time
ranges of the same duration. The first one was used as source to make predictions and the
second one was used to prove the predictions created previously. Due to the difficulty of
working with the statistics of individual tiles, the simplified model presented in Section 2 has
been used. Concretely, the experiment was conducted with the simplified model to the grid
cell defined by the level of resolution 12.
Figure5 shows the heatmaps of requests extracted from the web server logs of IDEE-Base
service, propagated to level 12 through the proposed model. These figures demonstrate that
some entities such as coast lines, cities and major roads are highly requested. These elements
could be used as entities for a predictive model to identify priority objects, as explained in [8].
These figures show that near levels are more related than distant ones, but all of them share
certain similarity. This relationships between resolution levels encourages the use of statistics
collected in a level to predict the map usage patterns in another level with detailer resolution.
For example, as shown in Figure5(c) and Figure5(e), resolution levels 14 and 16 are very
