Colocating mixed workloads from online-serving to offline-batch is a common way to improve resource utilization in modern data centers, which are increasingly heterogeneous in terms of CPUs and accelerators to meet diverse computing needs. A critical challenge in such environments is that workloads exhibit unique performance patterns on different hardware. For instance, while a scientific computing workload might run 1.8x faster on one hardware compared to another, a web-serving workload might show only a 1.1x improvement for the same hardware pair. These workload-specific performance variations cannot be captured by traditional metrics such as SPEC CPU scores, yet they significantly impact scheduling efficiency. Current schedulers overlook these complex workload-hardware performance relationships, resulting in inefficient resource allocation. To address this challenge, we propose MixSchedSim, a simulator to rapidly evaluate scheduling algorithms for mixed workload placement in heterogeneous computing environments. MixSchedSim explicitly models workload-specific performance variations across heterogeneous hardware, enabling the evaluation of scheduling algorithms that can exploit these patterns. The simulator employs an analytical model instead of detailed hardware simulation to achieve high simulation speed while maintaining accuracy. Using an event-driven approach to simulate state changes, MixSchedSim achieves linear-time complexity with respect to the product of the number of workloads and the cluster size, ensuring good scalability for large-scale evaluations. Our experimental validation using stress-ng CPU workloads shows that MixSchedSim achieves simulation accuracy of at least 66.6% compared to real-world scheduling results in a setup cluster. Using insights from MixSchedSim’s visualization capabilities, we developed a balanced scheduling algorithm that reduces makespan by up to 71.6% compared to existing approaches. MixSchedSim provides researchers and developers with a practical tool to develop more efficient scheduling algorithms through its easy-to-use YAML configuration interface, requiring no coding to run simulations. While the simulator has limitations such as requiring initial performance profiling and not modeling resource contention effects, it serves as the first simulator designed specifically for evaluating mixed workload scheduling in heterogeneous environments.