A Simulation Tool for Multiserver Job Queueing Systems

This thesis presents the Multiserver Job Queueing Model Simulator (MJQM-Simulator), a C++ tool built for high-performance analysis of Multiserver Job Queueing Models (MJQM). Queueing theory is essential for modelling, understanding and optimising performance in complex systems across various domains, allowing higher throughput at reduced response time. Systems may be studied either by analytical frameworks or by discrete event simulators. This thesis focuses on the latter approach. MJQMs have been recently introduced to study scheduling policies in large data centres especially after the demand explosion of AI workloads. A key feature of the MJQM-Simulator is its capacity to handle multiple job classes. Jobs can be differentiated by specific parameters, such as distinct arrival rates, resource demands and service time distributions, allowing for nuanced characterisation of heterogeneous workloads. Its modular architecture supports direct comparison of scheduling policies. Researchers can implement and test novel scheduling algorithms without modifying the simulation core. Human-readable TOML configuration files specify all parameters: system characteristics, per-class properties, and scheduling policy. Beyond reporting performance metrics such as server utilisation and mean queue lengths, the tool provides guided visual analysis that combines results from multiple experiments, supporting reproducible evaluation of multiserver systems under heterogeneous load conditions.