SmartCuckoo: A Fast and Cost-Efficient Hashing Index Scheme for Cloud Storage Systems

Fast query services are important to improve overall performance of large-scale storage systems when handling a large number of files. Open-addressing cuckoo hash schemes have been widely used to support query services due to the salient features of simplicity and ease of use. Conventional schemes are unfortunately inadequate to address the potential problem of having endless loops during item insertion, which degrades the query performance. To address the problem, we propose a costefficient cuckoo hashing scheme, named SmartCuckoo. The idea behind SmartCuckoo is to represent the hashing relationship as a directed pseudoforest and use it to track item placements for accurately predetermining the occurrence of endless loop. SmartCuckoo can efficiently predetermine insertion failures without paying a high cost of carrying out step-by-step probing. We have implemented SmartCuckoo in a large-scale cloud storage system. Extensive evaluations using three realworld traces and the YCSB benchmark demonstrate the efficiency and efficacy of SmartCuckoo. We have released the source code of SmartCuckoo for public use.