Things we learned about sums

QuestDB is an open-source time-series database designed for high-demand workloads, offering ultra-low latency and high ingestion throughput, while supporting Parquet and SQL to maintain data portability and AI readiness. In a recent release, QuestDB addressed concerns about numerical accuracy in summations by integrating compensated summation algorithms like Kahan and Neumaier, which correct errors that accumulate during large computations. This update allows QuestDB to compute accurate sums as quickly as naive ones, utilizing techniques such as prefetch and co-routines to optimize memory bandwidth. When compared to ClickHouse, QuestDB demonstrated significantly faster performance in summing large datasets with nulls. The database implements vectorized instructions to enhance the speed and accuracy of computations, providing consistent results across threaded operations and maintaining stability in aggregation tasks.