Evaluation and Simulation of Deferred Lock Acquisition and Deferred Lock Enforcement
Master thesis jointly supervised by L. Homann, G. Vossen WWU, and G. Graefe, Google
Database systems are about sharing structured information based on a schema (for a common interpretation of bits and bytes), physical data independence (to separate application logic from data representation), and transaction semantics. Transactions guarantee durability (persistence and availability, often by write-ahead logging) and concurrency control (often by locking). The goal of this thesis project is to simulate database locking and in particular the effects of controlled lock violation and of deferred lock enforcement on performance and scalability.
Controlled violation of locks held by a committing transaction permits other transactions to read updates before they are logged and durable. Thus, controlled lock violation is designed to increase concurrency while a transaction is hardening its committed updates. Deferred lock enforcement permits one transaction to read database records while another transaction updates them. Thus, it is designed to increase concurrency while transactions read and write database records and work through their application logic towards their commit points.
Our working hypothesis is that controlled lock violation and deferred lock enforcement complement each other. The objective of this simulation project is to refute or support this hypothesis and to quantify the techniques' effects on each other depending on database size, transaction size, read/write ratio, lock contention, etc. Crucial steps within this project include identifying a suitable simulation framework, simulating (coding) simple database transactions with and without these techniques, assessing the performance effects of the techniques, and preparing appropriate visualizations of simulation results.
The thesis provides a unique opportunity to collaborate with the 2017 ACM SIGMOD Edgar F. Codd Innovations Award Winner Goetz Graefe (Google). Goetz Graefe has an excellent reputation as database expert and is famous for his contributions to query optimization, query processing, and transaction processing. His current research focuses on concurrency control based on the two techniques mentioned above.