An OLS regression model for context-aware tile prefetching in a web map cache
An OLS regression model for context-aware tile prefetching in a web map cache
复制标题
DOI：
10.1080/13658816.2012.721555
复制
发表时间：
2013-03
影响因子：
5.7
通讯作者：
Ricardo García Martín;Juan Pablo de Castro Fernández;E. Verdú;M. Verdú;Luisa María Regueras Santos
中科院分区：
文献类型：
--
作者：
Ricardo García Martín;Juan Pablo de Castro Fernández;E. Verdú;M. Verdú;Luisa María Regueras Santos
随着网络地图服务的日益普及，空间数据基础设施中可扩展性更强的服务的开发也受到了推动。平铺地图服务已成为传统地图服务的可扩展替代方案。代替动态渲染地图图像，可以从服务器端缓存非常快速地提供预先生成的图像切片集合。然而，在服务启动期间，该高速缓存最初是空的，并且用户体验到较差的服务质量。图块预取尝试通过主动获取地图图像而无需等待客户端请求来提高命中率。虽然在传统的Web缓存中最流行的预取策略只考虑以前的访问历史来进行预测，但通过考虑背景地理信息，可以在Web映射中实现显着的改进。这项工作提出了一个回归模型来预测哪些地区可能会被要求在未来的基础上的空间互相关性之间的不受约束的目录的地理特征和过去的缓存请求的记录。可以预先生成并缓存预期最频繁请求的图块，以便更快地检索。跟踪驱动的模拟，从两个不同的全国范围内的公共网络地图服务在西班牙的数百万缓存请求表明，准确的预测和性能增益，可以获得与所提出的模型。
The increasing popularity of web map services has motivated the development of more scalable services in the spatial data infrastructures. Tiled map services have emerged as a scalable alternative to traditional map services. Instead of rendering map images on the fly, a collection of pre-generated image tiles can be served very fast from a server-side cache. However, during the start-up of the service, the cache is initially empty and users experience a poor quality of service. Tile prefetching attempts to improve hit rates by proactively fetching map images without waiting for client requests. While most popular prefetching policies in traditional web caching consider only the previous access history to make predictions, significant improvements could be achieved in web mapping by taking into account the background geographic information. This work proposes a regressive model to predict which areas are likely to be requested in the future based on spatial cross-correlation between an unconstrained catalog of geographic features and a record of past cache requests. Tiles that are anticipated to be most frequently requested can be pre-generated and cached for faster retrieval. Trace-driven simulations with several million cache requests from two different nation-wide public web map services in Spain demonstrate that accurate predictions and performance gains can be obtained with the proposed model.