HOTSPOT SENSITIVE DYNAMIC SCALING FOR DISTRIBUTED CACHE SYSTEMS

摘要

Distributed cache systems have been widely used to improve the applications' performance of accessing data by moving data and applications together in a cluster. However, existing works cannot well rebalance data partitions, and guarantee the data consistency and service availability during expanding cache systems. This paper proposes a hotspot sensitive dynamic scaling approach for distributed cache systems. First, we propose a hotspot sensitive data rebalance method, which is suitable for heterogeneous environment. It considers both memory utilization and network traffic, identifies hotspot data partitions, and then sets a high priority for the cache nodes with light workloads to guarantee the balance of cache nodes. Then, we propose a data accessing method based on two-phase controlled data migration to ensure the data consistency and service availability during scaling up cache systems. Finally, we have implemented a cache framework CacheScale, and conducted a series of experiments to validate the approach. The experimental results demonstrate that our approach can dynamically scale up cache systems, guarantee the data consistency and service availability, and achieve shorter response time.