**2. Tiling schemes**

Maps have been known for a long time only as printed on paper. Those printed cartographic maps were static representations limited to a fixed visualization scale with a certain Level Of Detail (LOD). However, with the development of digital maps, users can enlarge or reduce the visualized area by zooming operations, and the LOD is expected to be updated accordingly.

The adaptation of map content is strongly scale-dependent: A small-scale map contains less detailed information than a large scale map of the same area. The process of reducing the amount of data and adjusting the information to the given scale is called cartographic generalization, and it is usually carried out by the web map server [5].

In order to offer a tiled web map service, the web map server renders the map across a fixed set of scales through progressive generalization. Rendered map images are then divided into tiles, describing a tile pyramid as depicted in Figure 1.

**Figure 1.** Tile pyramid representation.

For example, Microsoft Bing Maps uses a tiling scheme where the first level allows representing the whole world in four tiles (2x2) of 256x256 pixels. The next level represents the whole world in 16 tiles (4x4) of 256x256 pixels and so on in powers of 4. A comprehensive study on tiling schemes can be found in [2].

## **2.1. Simplified model**

2 Will-be-set-by-IN-TECH

Most popular commercial services, like Google Maps, Yahoo Maps or Microsoft Virtual Earth, have already shown that significant performance improvements can be achieved by adopting

The potential of tiled map services is that map image tiles can be cached at any intermediate location between the client and the server, reducing the latency associated to the image generation process. Tile caches are usually deployed server-side, serving map image tiles concurrently to multiple users. Moreover, many mapping clients, like Google Earth or Nasa World Wind, have embedded caches, which can also reduce network congestion and network

This chapter deals with the algorithms that allow the optimization and management of these tile caches: population strategies (*seeding*), tile pre-fetching and cache replacement policies.

Maps have been known for a long time only as printed on paper. Those printed cartographic maps were static representations limited to a fixed visualization scale with a certain Level Of Detail (LOD). However, with the development of digital maps, users can enlarge or reduce the visualized area by zooming operations, and the LOD is expected to be updated accordingly. The adaptation of map content is strongly scale-dependent: A small-scale map contains less detailed information than a large scale map of the same area. The process of reducing the amount of data and adjusting the information to the given scale is called cartographic

In order to offer a tiled web map service, the web map server renders the map across a fixed set of scales through progressive generalization. Rendered map images are then divided into

generalization, and it is usually carried out by the web map server [5].

tiles, describing a tile pyramid as depicted in Figure 1.

**Figure 1.** Tile pyramid representation.

this methodology, using their custom tiling schemes.

delay.

**2. Tiling schemes**

Given the exponential nature of the scale pyramid, the resource consumption to store map tiles results often prohibitive for many providers when the cartography covers a wide geographic area for multiple scales. Consider for example that Google's BigTable, which contains the high-resolution satellite imagery of the world's surface as shown in Google Maps and Google Earth, contained approximately 70 terabytes of data in 2006 [6].

Besides the storage of map tiles, many caching systems also maintain metadata associated to each individual tile, such as the time when it was introduced into the cache, the last access to that object, or the number of times it has been requested. This information can then be used to improve the cache management; for example, when the cache is out of space, the LRU (*Least Recently Used*) replacement policy uses the last access time to discard the least recently used items first.

However, the space required to store the metadata associated to a given tile may only differ by two or three orders of magnitude to the one necessary to store the actual map image object. Therefore, it is not usually feasible to work with the statistics of individual tiles. To alleviate this problem, a simplified model has been proposed by different researchers. This model groups the statistics of adjacent tiles into a single object [7]. A grid is defined so all objects inside the same grid section are combined into a single one. The pyramidal structure of scales is therefore transformed in some way in a prism-like structure with the same number of items in all the scales.
