| In the last decade, the performance of CPUs is enhanced greatly, however, the performance growth of storage system cannot keep the pace, widening the performance gap between CPU and storage. The storage subsystem becomes a major bottleneck of the overall computer system, leading to low computer resource usage and high production cost. High performance flash memory based SSDs are adopted to replace traditional disks to mitigate this situation. Due to its high price, SSDs are generally used as cache of slow storage devices in commercial environment.On the other hand, Virtualization technology, as the core of cloud computing, enables one physical machine to host multiple virtual machines simultaneously, which greatly improves hardware resource usage and reduces cost. In general, large similarities exist among virtual machine images as they run similar operating systems, libraries and applications. However, in actual production environment, each virtual machine is assigned distinct cache space from the shared cache device. Data duplication among virtual machines is ignored and not considered, hurting the utility of cache device substantially.In order to solve this problem, this paper designs and implements a cache deduplication system for SSD based cache device. The main method is to cache hot data on SSDs and introduce deduplication mechanism to increase cache usage. What’s more, we design a Dedu-ARC cache replacement policy, which is deduplicate-aware and based on ARC algorithm. Experiment results show that, compared to traditional cache system, L-CACHE can improve the cache hit ratio and reduce average I/O latency. The main work and contributions of this paper are:1. Analyze the design and implementation of general block layer and Device Mapper in Linux kernel.2. Design and implement L-CACHE system which is a deduplication aware cache.3. Design and implement Dedu-ARC cache replacement policy in L-CACHE system.4. Present an evaluation study of L-CACHE system. |
PDF Full Text Request