Adaptive indexing approach for main memory column store

Abstract

Owing to efficient query processing for random workload, the hybrid crack sort (HCS) has become an important adaptive indexing approach in main-memory column store. However, under sequential workload scenarios, the HCS does not obtain a good query execution performance, because of great reorganisation overhead imposed on the initial queries. The authors propose a hybrid radix crack sort (HRCS) approach to solve this problem. By the adoption of radix-based partition strategy, it divides the unsorted column into disjoint key ranges and then conducts data reorganisation in at most two key ranges for each query. For HRCS, only a small portion of the whole column needs to be touched for the processing of each query, thus reducing the reorganisation cost and improving the query execution performance. The final experiments show that the novel HRCS approach can obtain a higher query execution performance for not only random workload but also sequential workload, as compared with HCS.