Adaptive Hybrid Indexes

Database systems leverage index structures to improve query performance. However, indexes can also significantly increase a database's memory footprint, making them an ideal compression target. Nevertheless, there is a trade-off between space and performance as more compressed encodings often result in decreased query performance. We, therefore, propose a framework called Adaptive Hybrid Indexes (AHI) that allows optimizing indexes adaptively at run-time at a finer granularity by considering the workload properties. Instead of representing all index parts using one encoding, AHI can implement different encodings which are adaptively applied depending on the current workload. For example, frequently accessed parts could be stored in performance-optimized encodings, while rarely accessed parts could be represented in a more compressed fashion. Our experiments show that this approach can reduce index sizes significantly while retaining most of the performance under skewed workloads.