中文摘要：

随着云计算的普及,软件即服务（software as a service,SaaS）逐渐成为云计算的一种重要表现形式.云中数据节点的缓存是提高多租户应用数据访问性能的一种重要资源,缓存资源的共享和分配受到SaaS提供商的关注.对SaaS提供商而言,如何在多租户间有效地分配数据节点上的缓存资源,从而满足租户的服务水平协议（service level agreement,SLA）,获得更高的收益已成为一项挑战.为此,提出了多租户云数据存储缓存管理机制,以实现服务提供商收益最大化的目标,结合SLA收益模型,评估不同缓存策略下服务提供商获取的收益值,将全局缓存管理问题定义为目标优化问题,并结合缓存分配特点,采用优化的遗传算法解决该问题.通过实验比较,该方法能保证SaaS服务提供商在多租户间有效利用缓存资源获取高收益.

英文摘要：

With the popularization of cloud computing, software as a service （SaaS） has become an important form of cloud computing. Memory resource owned by each data node in the cloud is a key resource to improve data access performance of multi-tenant applications. Therefore, memory resource share and provisioning have received a lot of attention from SaaS providers. For the service providers, how to reasonably allocate memory resource in each data node in order to obtain higher profits while guaranteeing tenants＇ service level agreement （SLA） has become a challenge. Addressing the challenge, we propose a framework of multi-tenant memory management （MTMM） for cloud data storage and corresponding memory allocation method. The method takes the maximum profits service provider can obtain as a target. Combined with tenants＇ SLA profit models, the global memory allocation problem is analyzed and modeled as an objective optimal problem. Corresponding the profits service provider can get under different memory allocation strategies are predicted through it. Considering the characteristics of multi-tenant memory allocation, we solve the problem by optimized genetic algorithm in order to improve the performance of the method. Compared with the traditional LRU method and multi-tenant memory allocation method employed in single node, the mechanism proposed in this paper can effectively manage memory and provide higher profits for service providers.