When Can We Trust Progress Estimators for SQL Queries?

The problem of estimating progress for long-running queries has recently been introduced. We analyze the characteristics of the progress estimation problem, from the perspective of providing robust, worst-case guarantees. Our first result is that in the worst case, no progress estimation algorithm can yield anything even moderately better than the trivial guarantee that identifies the progress as lying between 0% and 100%. In such cases, we introduce an estimator that can optimally bound the error. By placing different types of restrictions on the data and query characteristics, we show that it is possible to design effective progress estimators with small error bounds. We show where previous solutions lie in this spectrum. We then demonstrate empirically that these "good" scenarios are common in practice and discuss possible ways of combining the estimators.

• Type: Conference
• Year: 2005
• Where: SIGMOD
• Authors: Surajit Chaudhuri, Raghav Kaushik, Ravishankar Ramamurthy
• Keywords: Database, Effective Progress Estimators, Progress Estimation Algorithm, Progress Estimation Problem, SIGMOD 2005