Table of Links

Abstract and 1 Introduction

1.1 Configuration Parameter Tuning Challenges and 1.2 Contributions

2 Tuning Objectives

3 Overview of Tuning Framework

4 Workload Characterization and 4.1 Query-level Characterization

4.2 Runtime-based Characterization

5 Feature Pruning and 5.1 Workload-level Pruning

5.2 Configuration-level Pruning

5.3 Summary

6 Knowledge from Experience

7 Configuration Recommendation and 7.1 Bayesian Optimization

7.2 Neural Network

7.3 Reinforcement Learning

7.4 Search-based Solutions

8 Experimental Setting

9 Related Work

10 Discussion and Conclusion, and References

2 TUNING OBJECTIVES

In the context of DBMS configration uning, the primary objective is to maximize the Performance, such as throughput and latency (e.g., 95th percentile latency). This aims to gauge the DBMS’s ability to efficiently handle a higher volume of workloads or queries.

Additionally, several other aspects/objectives can also be considered during the tuning process as summarised in Table 1, including:

• Overhead: Focuses on the amount of time or system resources the method requires to recommend knob settings, including tuning time, Total Cost of Ownership (TCO), and resource utilization (such as CPU utilization), etc. For example, RelM [19] concentrates on optimizing memory allocation for such applications. RestTune [6] aims to reduce resource utilization while still guaranteeing the Service Level Agreement (SLA), e.g., without violating throughput and latency requirements.

• Adaptivity: Evaluates how well the method performs in new and varying scenarios, such as hardware adaptation for changes in I/O, memory, workload adaptation for varying request rates and data sizes, etc.

• Safety: One crucial facet of safety lies in the ability of a tuning method to avoid recommending parameter configurations that could potentially degrade system performance. ONLINETUNE [3], which focuses on tuning online databases safely in changing cloud environments, addresses this concern. The safety threshold, as defined in ONLINETUNE, is anchored to the baseline performance exhibited under the default configuration, denoted as default performance. Beyond performance-centric safety assessments, resource allocation represents another pivotal dimension. Within this context, Kunjir et al. [19] present RelM, wherein safety concerns are framed in terms of resource utilization aligning with allocated thresholds