SIGMOD 2019 Highlights: Blockchain Keynote, CDBTune Paper on Cloud Database Tuning, and Emerging Database Research

ACM SIGMOD/PODS 2019 was held from June 30 to July 5 in Amsterdam, and the Tencent technology team reported on‑site about the conference highlights.

The first day keynote was titled “Responsible Data Science”; the second day featured a blockchain‑focused keynote by IBM Almaden researcher C. Mohan, discussing private blockchain systems and comparing them with traditional databases.

On the second day, a joint paper by Huazhong University of Science and Technology and Tencent TEG Cloud Architecture Platform CDB team, titled “An End‑to‑End Automatic Cloud Database Tuning System Using Deep Reinforcement Learning” (CDBTune), was presented as the first paper in the Distributed Data Management track.

Parameter tuning is crucial for cloud database performance, but existing methods suffer from high cost, need for large high‑quality samples, difficulty handling high‑dimensional continuous parameters, and limited adaptability to cloud environments.

CDBTune addresses these challenges with an end‑to‑end system that employs deep reinforcement learning (DDPG) to search for optimal configurations in a high‑dimensional continuous space; the overall architecture is illustrated below.

CDBTune uses a try‑and‑error strategy, initializes training with a small sample set, and replaces traditional regression with a reward‑feedback mechanism to accelerate convergence.

Experimental results show that CDBTune achieves strong performance improvements and good generalization capability.

The paper title and a link to the original ACM record are provided: An End‑to‑End Automatic Cloud Database Tuning System Using Deep Reinforcement Learning .

The report also highlights the “Modern Hardware” theme, featuring six papers on multi‑core logging and recovery, NUMA dataflow processing, RDMA‑based index design, GPU acceleration for matrix and graph workloads, and other emerging hardware impacts on database research.

Additional academic topics covered at the conference include query processing and optimization, data provenance, stream processing, data integration/cleaning, and graph processing, with three dedicated graph sessions indicating the field’s popularity.

The System Award was presented to Amazon’s Aurora team for its innovative cloud‑native relational database storage design.

In the indexing track, IBM presented a succinct secondary indexing mechanism exploiting column correlations, while Microsoft showcased an AI‑driven approach to improve index recommendations, both of which are readily integrable into commercial database products.

A data‑platform paper titled “Socrates: The New SQL Server in the Cloud” described an architecture similar to Aurora’s but with a separate logical logging layer, offering valuable design insights.

During the conference, Tencent’s “Rhino Night” event gathered scholars to discuss hot technical topics and explore industry‑academic collaboration models.

Tencent Database Technology Team supports a range of products such as CDB, CTSDB, TDSDB, CKV, CMonGo, and continuously enhances core database capabilities, improves performance, ensures system stability, and solves operational issues in production environments.