View article

Efficient utilization of dynamic random access memory (DRAM) is crucial for achieving high-performance query processing in database systems, especially as data volumes continue to grow. Unfortunately, the cost of DRAM is unlikely to decrease in the coming years, and it is already the dominating cost factor in modern data centers. Consequently, lightweight in-memory compression techniques can reduce the memory footprint and maximize the data stored in memory. However, compressing all data, regardless of the compression algorithm’s efficiency, causes additional CPU overhead during query execution. To address this challenge, we introduce AdaCom, a novel framework that selectively applies lightweight succinct encodings only to infrequently accessed data. By doing so, we mitigate the performance overhead associated with compression. In our experimental evaluation, we demonstrate that AdaCom reduces the memory footprint by up to 40% while retaining most of the performance (≈ 95%).