Toward digital twins' workload allocation on clouds with low-cost microservices streaming interaction

A Digital Twin (DT) is a set of computational models representing real-time physical objects and processes in a digital world. The increasing adoption of this paradigm by the major industrial equipment vendors to simulate real-time working conditions and perform smart decision-making, established the Smart Factory Digital Twins architectures, where a set of DTs published as microservices interact with each other exchanging information by streaming technology. In this sense, a fundamental problem consists of selecting adequate computational resources to simulate the physical objects. In an Infrastructure as a Service (IaaS) cloud for DTs, the allocation focuses on distributing the jobs of the set into the virtual machine instances in a way that computational resources demand is satisfied and the cost is minimized. In this paper, we propose a set of algorithms based on heuristics, metaheuristics, and Mixed Integer Programming to find low-cost solutions. The performance of algorithms is evaluated using Amazon EC2 instances and DT jobs with randomly generated bandwidth, memory, and processor requirements. The experimental results show that the proposed approaches based on bin packing, genetic algorithms, partition, filtering, set coverage, and branch and bound strategies present a competitive performance in the workload allocation of the computational set of jobs of a DT into an IaaS cloud environment. Our allocation heuristic-based techniques allow considerable cost savings in medium and large periods concerning standard approaches such as local search.