Associativity-Aware Transaction Processing Optimization for Web Applications

Web applications rely on database systems to store and manage data. Existing major databases use lock-based concurrency control to coordinate transactions. However, lock-based transaction processing can be expensive and result in unnecessary deadlocks because the database systems are oblivious to the application semantics, such as access patterns. This paper examines an issue commonly seen in web applications: inefficient associated accesses. They are problematic because they cause database systems to abort concurrent transactions frequently, degrading application performance. To address this issue, we build Railyzer, a tool that automatically analyzes web applications’ database access patterns and provides possible optimization strategies. First, we establish a criterion to differentiate associated accesses from other accesses. Then, based on a language-agnostic analysis of the application, we create a method for identifying associated accesses. Finally, we use heuristics to locate inefficient ones and suggest fixes. We discovered 83 potential optimizations among the six open source applications. Some of these optimizations improve the application throughput by up to 70%.