Dynamic workload management for very large data warehouses: juggling feathers and bowling balls

摘要: Workload management for business intelligence (BI) queries poses different challenges than those addressed in the online transaction processing (OLTP) context. The fundamental problem is that execution times of BI can range from milliseconds to hours, and it difficult estimate these accurately. Key raised by this are how identify not performing properly what do about them. We propose here a workload system controlling individual based on realistic customer service level objectives. In order validate our proposal, we have implemented an experimental includes dynamic controller leverages fuzzy logic. We present results number experiments ran using workloads actual industrial objectives gathered interviewing industry practitioners. Our show even handful moderately mis-behaving significant impact consisting thousands queries. were surprised when also demonstrated false positives -- incorrectly identifying normal query as consequences. For reasons, very important be accurate possible avoiding both negatives. Our markedly improve