基于非均衡时空访问模式的网络地图服务计算资源弹性分配方法

The spatiotemporal imbalance and the dynamic features of user access behaviors make the server-side computing resource allocation for web map services challenging. Inappropriate allocation leads to the imbalances of provision-and-demand, and it will then significantly impact the quality of service and also lead to the waste of computing resources. The following factors make the optimal allocation even more challenging, such as the difference on access contents, the variety on the functions and the capacities of web map services, the competition of user access and the limited resources. This research project aims at: 1) systematically analyzing the spatiotemporal access patterns of web map services, and quantifying the correlation between access patterns and computing resource demands; 2) establishing the metrics and evaluation criteria for measuring the balancing of provision-and-demand of computing resources on the basis of quality of geospatial information services (QoGIS), access coverage and the utilization rate of computing resources; 3) proposing an elastic computing resource allocation method combined with constrained dynamic server site selection to promote the overall quality of global users. This method will takes user access competitions from dispersed locations and the constraints on resource allocation into account, by learning from site selection and Nash Equilibrium in game theory. It is expected that, through this research, the further understanding of the imbalance spatiotemporal access patterns and the influences on the demands of computing resources will be clarified. The methodologies proposed in this project will enrich the quality evaluation and optimization methods of geospatial information services. The significance underlying this research is that the potential outcome can be used as the guideline for advising the optimal deployment of geospatial information services in distributed environments.

用户访问行为的时空非均衡性和高动态性，对网络地图服务的服务端计算资源动态分配提出了很高的要求。不合理的计算资源分配将导致供求失衡，既影响地图服务质量，又浪费计算资源。地图访问内容的差异性、服务功能的多样性、用户访问竞争和资源使用受限等因素，进一步增加了计算资源优化分配的难度。本项目拟，1）系统分析地图服务时空访问模式，对访问模式与计算资源需求之间的关联关系量化建模；2）基于空间信息服务质量、访问覆盖与计算资源利用率度量，建立计算资源供求时空匹配度的指标体系和评价准则；3）借鉴空间选址的思想并融合纳什均衡理论，提出一种顾及多点用户访问竞争和资源分配约束的地图服务有限离散时空动态选址及计算资源弹性分配方法，从整体上提升地图服务质量。本项目研究将阐明非均衡时空访问模式及其对计算资源供求的影响，丰富空间信息网络服务评价与优化的方法，并对分布式环境下实施空间信息服务应用的优化部署具有重要的指导意义。

针对在线网络地图服务（WMS）质量参差不齐，用户并发访问下性能低下的问题，本项目以WMS“在线爬取-质量分析-性能优化-检索改善”为主线，研究基于非均衡用户时空访问规律的WMS性能优化和WMS资源高效准确检索的关键理论方法与技术。具体成果如下：.1）设计并实现了一种基于最佳优先搜索和地理信息资源描述特征自学习的WMS主题爬取算法，获取全球范围内4万余WMS；并从WMS的不同维度开展了统计分析，为地图时空访问模式建模、性能改善提供和检索优化提供了充足的样本数据和依据。.2）提出了一种弹性可配置的、高时空覆盖的分布式监测框架，挖掘了WMS服务质量，特别是响应时间的时空变化规律，进而提出了一种顾及用户访问时空分异特征的GWR与STARMA融合的WMS响应时间时空预测模型，实现WMS的用户时空访问性能的精准评估。.3）基于公有云平台设计了一种全球式分布式弹性伸缩部署架构及算法，优化了分布式地理信息门户的负载均衡和响应时间，解决了用户时空并发访问不均情况下的计算资源动态调配和服务访问加速问题。.4）融合SWEET本体、WordNet词库和语义相似度距离度量，提出了一种地理语义驱动的地学领域主题多标签分类算法和元数据扩展描述机制，解决无法按地学领域主题检索的问题；设计了一种多核多特征融合学习和用户反馈结合的WMS图层检索方法，利用用户地图视觉认知动态调整和精化检索规则，应对WMS地图图文不符、用户查询需求多样且难以表述的问题，提升了WMS检索的精度与效率。.5）开发了性能监测分析与检索可视化系统原型，验证了上述研究成果的有效性。.本项目的研究成果可用于指导地理信息服务性能的监测、评估与优化，及不同类型在线地理信息资源的检索与发现，特别是对于地理信息门户网站的设计与优化具有重要的指导意义。