Three Papers from OceanBase Accepted at ICDE 2024

From May 13 to 18, 2024, the annual IEEE International Conference on Data Engineering (ICDE 2024) was held in Utrecht, the Netherlands. Three papers co-authored by research teams from colleges, OceanBase, and other partners were accepted and presented at the conference.

ICDE is one of the premier conferences in data and information engineering and addresses research issues on designing, building, managing, and evaluating advanced data intensive systems and applications. It is a leading forum for researchers, practitioners, developers, and users to explore cutting-edge ideas and to exchange techniques, tools, and experiences.

The acceptance of the ICDE 2024 papers marks a further breakthrough in scientific research and innovation of OceanBase after the deadlock detection paper was accepted at ICDE 2023. Chuanhui Yang, CTO of OceanBase, delivered a speech titled "OceanBase: From OLTP to HTAP" at ICDE 2024, addressing the technical challenges encountered in the evolution of OceanBase from distributed OLTP to HTAP real-time analytics, as well as the latest progresses in column store, resource isolation, complex queries, and multi-model of distributed SQL databases.

Accepted Paper 1: Functionality-Aware Database Tuning via Multi-Task Learning

The accepted paper "Functionality-Aware Database Tuning via Multi-Task Learning" was jointly published by East China Normal University and OceanBase. In order to enhance the knobs tuning results in OceanBase and other databases, this paper proposes a functionality-aware database knobs tuning framework named OBTune, and verifies its generalizability on various databases.

The main findings of the research are as follows:

1. The paper demonstrates the problems that may be caused by existing black box tuning methods when irrelevant knobs are being tuned simultaneously and indicates the potential risks of applying improper parameters to online database instances.

2. The paper proposes a functionality-aware database tuning framework based on multi-task learning, which enhances the tuning results through learning the relationships between different tasks, and avoids adjusting irrelevant konbs by perceiving the status of functionalities. The framework takes the database's overall performance as the objective of main learning task, and each function module as a separate learning task.

3. The functionality-aware automatic tuning method (i.e., OBTune) is verified on the distributed database, OceanBase, and the centralized database, PostgreSQL. Experimental results show that better performances were achieved in knobs tuning on stand-alone databases and distributed databases (stand-alone distributed integrated databases).

ICDE reviewer gave a high evaluation of this paper and the innovative database tuning framework it proposed, and added that the framework showed its advantages and potential to the industry with a detailed design and implementation, while promoting the development of existing technologies.

Selected Paper 2: Generative and Contrastive Paradigms Are Complementary for Graph Self-Supervised Learning

The research paper authored by Wuhan University, OceanBase, and patners proposes the graph contrastive masked autoencoder (GCMAE) framework, which combines the generative and comparative learning paradigms to improve the performance of graph self-supervised learning. In existing graph self-supervised learning tasks, masked autoencoder (MAE) and contrastive learning (CL) are used separately. But the authors observe that they are complementary and propose the GCMAE. The framework integrates the MAE and the CL branches, which share a common encoder, so that GCMAE can make full use of global information. The team evaluated GCMAE on four popular graph tasks and compared it with 14 state-of-the-art baselines. The results show that GCMAE consistently provides good accuracy across tasks, with a maximum accuracy improvement of up to 3.2% over the best-performing baseline.

Accepted Paper 3: BENCHTEMP: A General Benchmark for Evaluating Temporal Graph Neural Networks

The paper presented by Wuhan University, ETH, eBay, and OceanBase introduces BecnchTemp, a general benchmark for evaluating temporal graph neural network (TGNN) models on various workloads. BenchTemp provides a set of benchmark datasets and a standard pipeline that unifies the TGNN evaluation so that researchers can compare different TGNN models quickly and efficiently. BecnchTemp supports multiple tasks and settings, and examines the effectiveness and efficiency of the models at the same time.

Up to now, OceanBase has published more than 20 papers in prestigious academic journals, including SIGMOD, ICDE, VLDB, etc. The technical capabilities and innovative contributions of OceanBase have been recognized by the academic community in the database industry around the world.