Joins on high-bandwidth memory: a new level in the memory hierarchy

High-bandwidth memory (HBM) gives an additional opportunity for hardware performance benefits. The high available bandwidth compared to regular DRAM allows execution of many threads in parallel, avoiding memory stalls through many concurrent memory accesses This is especially interesting considering database join algorithms optimized for multicore CPUs, even more when running on a manycore processor like a Xeon Phi Knights Landing (KNL). The drawback of HBM, however, is its small capacity as well as under-utilization in random memory access patterns. In this paper, we analyze the impact of HBM on join processing on the KNL architecture. We evaluate main memory hash join and sort-merge join algorithms of relational DBMS as well as data stream joins, comparing execution times in different HBM configurations. Our results show performance gains up to 3 × for joins when HBM is used. Finally, we summarize our lessons learned, give additional advice for HBM utilization, and discuss generalizations for other levels of the memory hierarchy.