CSE
CSE
CSE CSE


Defense Event

Autonomous Database Management at Scale: Automated Tuning, Performance Diagnosis, and Resource Decentralization

Dong Young Yoon


 
Friday, August 23, 2019
09:00am - 11:00am
3725 Beyster Building

Add to Google Calendar

About the Event

Database management systems are one of the most mission-critical software components for enterprises. However, its ever-increasing complexity constantly raises its operating costs. To reduce such costs, many enterprises have outsourced databases to Database-as- a-Service (DBaaS) in the cloud in recent years. Today, it is very crucial for cloud service providers and database vendors to be able to maintain a myriad of databases operating simultaneously. Given databases’ ever-increasing complexity and scale, it is becoming less feasible for database administrators (DBAs)—who are already very costly and scarce resources—to maintain and optimize each database day-and-night manually. Because of this, the notion of autonomous databases has gained much attention from the database research community with a goal to make databases operating at full capacity with zero human intervention. Previous research on autonomous database management systems has been mainly limited to a few specific topics, namely physical design and parameter tuning. Limited research has been done on new technologies in the context of autonomous databases, such as distributed databases and approximate query processing (AQP). This dissertation discusses three specific areas of modern database systems that can improve to achieve better automation of current databases: 1. Automated Tuning – We empirically study: a) a new type of database designer that is robust against uncertain future workloads; and b) a hybrid sampling scheme with an optimal strategy for joins on samples in approximate query processing (AQP). 2. Performance Diagnosis – We present a practical tool for assisting users in quickly and reliably diagnosing performance problems in transactional databases. 3. Resource Decentralization – We propose a novel distributed lock manager that can provide a full resource decentralization in distributed database systems.

Additional Information

Sponsor(s): Barzan Mozafari

Open to: Public