Multi-server systems, where multiple servers collaborate to provide various services, are integral to numerous real-world applications, such as load balancing in cloud computing, packet scheduling in wireless networks, and parallel processing in scientific computing. These systems are crucial for enhancing the efficiency and reliability of communication and computer systems, as they allow for optimized resource allocation and management in dynamic and complex environments. This research addresses the challenging problem of job scheduling in heterogeneous multi-server systems, where different queues correspond to distinct job types, each with unique Quality of Service (QoS) requirements. The complexity of this problem stems from the dynamic nature of the environment, the heterogeneous characteristics of the servers, the varying arrival rates of jobs into different queues, and the need to balance performance and cost. Efficient job scheduling is essential to maximize system performance and minimize operational costs, but the dynamic and multi-queue nature of these systems makes it a highly complex problem to solve. To tackle this issue, in this research, we propose two novel algorithms: Load and Deadline Aware (LDA) and Minimum Percentage of Violations (MinPoV). The LDA algorithm dynamically allocates jobs by integrating load metrics with deadline considerations, ensuring a balanced distribution of jobs. On the other hand, the MinPoV algorithm combines MaxWeight and Earliest Deadline First (EDF) strategies to minimize deadline violations, thus enhancing system efficiency. Extensive experiments were conducted to evaluate the performance of the proposed algorithms against existing benchmarks such as Random, Round Robin, First Come First Serve (FCFS), and MaxWeight. The evaluation metrics included Average Response Time, Average Waiting Time, and Average Percentage of Violations. The results demonstrate that the proposed algorithms significantly outperform the baseline methods across all metrics. Notably, the MinPoV algorithm exhibits superior performance compared to LDA, particularly in minimizing deadline violations. These findings suggest that both LDA and MinPoV are promising solutions for job scheduling in heterogeneous multi-server systems with dynamic and multi-queue environments, offering significant improvements in performance and reliability .
