Experimental Investigation of Database Behavior Under Progressive Concurrent Workloads

This research presents an experimental investigation of database behavior under progressively increasing concurrent workloads using SQLite 3.45.1. The study evaluates concurrency levels from 1 to 128 client processes under a controlled read/write workload using Write-Ahead Logging (WAL). The experiment measures transaction throughput, latency, lock conflicts, resource utilization, fairness, transaction aborts, and scalability as the number of concurrent clients increases. Each client operates as an independent process and database connection using a closed-loop workload consisting of 80% read transactions and 20% write transactions with a Zipfian access distribution. The results identify an operational saturation point using a predefined 5% saturation criterion. The measured peak mean throughput occurs at 32 concurrent clients, while 16 clients are identified as the operational saturation point under the defined criterion. At higher concurrency levels, particularly 128 clients, the experiments show increased latency, lock contention, memory consumption, and greater performance variability. This repository contains the experimental implementation, configuration, raw and processed measurements, statistical analysis, generated figures, and instructions required to reproduce the experiments. The work is intended to provide an empirical basis for understanding concurrency saturation behavior in SQLite under controlled progressive workloads.