Geography Reference
In-Depth Information
correlated. It is easier to work with the statistics of level 14 than with those of level 16 which
has much more elements.
Table 2 represents the hit percentage achieved by using this model for the IDEE-Base service.
This table shows the percentage of hits obtained for the level identified by the column index
from the statistics collected in the level identified by the row index. Last column shows
the resources consumption, as a percentage of cached tiles. Last row collects the results of
combining the statistics of all levels to make predictions over every level. Shadowed cell in
Table 2 indicates that using retrieved statistics of level 13 as the prediction source, a hit rate
of 92.1573% is obtained for predictions made in the level 18, being necessary the storage of a
25.8049% of the tiles in cache.
12 13 14 15 16 17 18 19 resources
12 98.6417 98.9362 99.3573 99.4737 99.6901 99.2637 99.2993 94.7561 40.2172
13 87.8163 93.5760 95.8939 96.4146 97.4372 95.2686 92.1573 75.5073 25.8049
14 53.0529 61.5825 86.7783 88.2709 91.1807 81.6460 63.4527 43.9129 9.3302
15 37.1553 47.9419 77.9136 84.0861 83.7095 69.9489 57.0746 33.3348 5.2354
16 46.9387 57.5640 84.3110 86.7747 91.8272 78.7670 64.3781 41.8433 7.7686
17 30.2021 37.6348 57.0138 60.1330 62.2834 69.5106 55.5134 23.4647 3.2676
18 23.5791 25.5913 41.7535 46.1559 45.8693 41.4502 61.9763 33.3799 2.3291
19 8.8690 8.6848 12.4556 13.1338 14.3302 12.2756 13.6932 44.1113 1.2295
prop 98.9340 99.3080 99.6074 99.6321 99.7763 99.4244 99.4308 97.2315 41.3647
Table 2.
Percentage (%) of cache hits through the simplified model obtained from
IDEE-BASE
logs,
using the mean of the normalized frequencies as the probability threshold.
Nevertheless, it must be noted that the main benefit of using a partial cache is not the
reduction in the number of cached tiles. The main benefits are the savings in storage space
and generation time. As explained in [8], the amount of saved tiles is bigger than the storage
saving. It reveals that the most interesting tiles come at a bigger cost. Mainly, popular areas
are more complex, and it is necessary more disk space to store them.
Figure6 and Figure7 represent the cache hit ratios obtained by the simplified model for the
IDEE-BASE
service. This model bases its operation on the knowledge of past accesses,
assuming a certain stationarity of requests; it assumes that map regions that have been
popular in the past will maintain its popularity in the future. However, from a certain
percentage of cached objects, identified by the continuous vertical line, the simplified model
is not able to make predictions. Tiles situated at the right of this line correspond to objects that
have never been requested so are not collected in server logs. To complete the model, these
never-requested tiles have been randomly selected for caching, yielding a linear curve for this
interval.
Results demonstrate that the simplified model obtains better results for predicting user
behavior from near resolution levels. For low-resolution levels high cache hit ratios are
achieved by using a reduced subset of the total tiles. However, descending in the scale
pyramid, the requested objects percentage decreases, so the model prediction range and its
ability to make predictions decrease too. For future work, instead of randomly selecting
objects for caching in this interval, interesting features could be identified and used to define
priority objects.
