Capacity Aware Container Placement in Heterogeneous Clusters using Genetic Algorithm

High performance application execution in a containerized environment has been a popular field of research in Cloud Computing. With Docker containers being lightweight and providing complete isolation, executing tasks with heavy resourcedemands has become effortless. Due to the ubiquity of containers,highly heterogeneous and unpredictable workloads proceed that sometimes result in resource exhaustion leading to start-up latency which can become a hindrance in the execution process. As the resource requirements from these diverse workloads cannot be always matched accurately, there is a probability for resource contention, which limits the scaling levels for further task-based containers and generates a significant delay. Therefore, there persists a trade-off between resource utilization and latency. Despite prevalent research work being done in this area, there still remains some open issues in the field of optimized container allocation in the cluster. In this research, Genetic Algorithm is used to generate a capacity-aware container placement method to ensure there is a significant amount of resource available to service mthe deployed application tasks with different resource requirements by creating a fitness function based on the capacity threshold parameter. On implementation, it is observed that there is a 40% increase in the resource utilization level, with 5000 seconds of improvement in execution time. On comparing the individual utilization values with the average utilization of the cluster, a balanced cluster is seen to be achieved