The aim of this Dagstuhl Seminar on Robust Query Processing is to bring together top researchers from both industry and academia working on plan generation and plan execution in database query processing and in cloud-based massively parallel systems for data management. Delivering robust query performance is well known to be a difficult problem for database management systems. All experienced DBAs and database users are familiar with sudden disruptions in data centers due to poor performance of queries that have performed perfectly well in the past. The goal of the seminar is to discuss the current state-of-the-art, to identify specific research opportunities in order to improve the state-of-affairs in query processing, and to develop new approaches or even solutions for these opportunities.

Prior Dagstuhl Seminars have initiated discussions on robust performance of query processing in relational databases. They have proposed some metrics and benchmarks as well as initial solutions. These prior solutions have focused on data access, e.g., index scans vs. table scans as well as dynamic transitions from one to the other, and on execution algorithms, e.g., for joins and duplicate removal. Much work remains, e.g., on join sequences and on load balancing. It could be argued that the remaining work will be harder than the prior work – therefore, we invite to this Dagstuhl Seminar on Robust Query Processing.

The seminar will start with a focus on the problem and on prior results. Thereafter, seminar participants will work in small groups on specific problems. For example, one group might focus on detecting that a chosen join sequence is suboptimal for the actual predicate selectivities whereas another group might focus on changing dynamically the sequence of join and aggregation operations. Perhaps multiple groups focus on changing the join sequence assuming alternative join algorithms, e.g., order-based merge joins or order-agnostic hash joins. Interleaved plenary presentations and discussions will re-focus the working groups. Towards the end of the week, we may have concrete ideas for new techniques and perhaps for publications. Both academic and industrial participants may freely use the discussion contents and results for follow-on work.

Creative Commons BY 3.0 DE

Renata Borovica, Goetz Graefe, and Allison Lee

• Angelos Christos Anadiotis (EPFL - Lausanne, CH) [dblp]
• Tahir Azim (EPFL - Lausanne, CH) [dblp]
• Peter A. Boncz (CWI - Amsterdam, NL) [dblp]
• Renata Borovica-Gajic (The University of Melbourne, AU) [dblp]
• Surajit Chaudhuri (Microsoft Research - Redmond, US) [dblp]
• John Cieslewicz (Google Mountain View, US) [dblp]
• Thanh Do (Google - Madison, US) [dblp]
• Campbell Fraser (Google - Kirkland, US) [dblp]
• Johann-Christoph Freytag (HU Berlin, DE) [dblp]
• Goetz Graefe (Google - Madison, US) [dblp]
• Fabian Hüske (data Artisans - Berlin, DE) [dblp]
• Alfons Kemper (TU München, DE) [dblp]
• Allison Lee (Snowflake Computing Inc. - San Mateo, US) [dblp]
• Thomas Neumann (TU München, DE) [dblp]
• Anisoara Nica (SAP SE - Waterloo, CA) [dblp]
• Glenn Paulley (SAP SE - Waterloo, CA) [dblp]
• Ilia Petrov (Hochschule Reutlingen, DE) [dblp]
• Bart Samwel (Google - Amsterdam, NL) [dblp]
• Kai-Uwe Sattler (TU Ilmenau, DE) [dblp]
• Knut Stolze (IBM Deutschland - Böblingen, DE) [dblp]
• Immanuel Trummer (Cornell University - Ithaca, US) [dblp]
• Srinivas Karthik Venkatesh (Indian Institute of Science - Bangalore, IN) [dblp]
• Jiaqi Yan (Snowflake Computing Inc. - San Mateo, US) [dblp]
• Marcin Zukowski (Snowflake Computing Inc. - San Mateo, US) [dblp]