Learning-based Approach on Mathematical Models of Microservices for Various Optimization

Adoption of microservices has increased exponentially worldwide.Despite higher adoption, there is a lack of work on possible auto tuning of configurable parameters for various optimisation in terms of cost, resources and other aspects.We propose a novel way to find the best possible configuration for a service based on certain parameters.Each service is modeled mathematically by considering the various configurable parameters that a service can have.We then use learning methodologies to optimise on the cost and find the best parameters for optimised on the cost without affecting the service, thereby cutting down on over provisioning of the configuration.Our in-house implementation and experimentation on Samsung owned microservices achieved an optimization in terms of cost by $1593 per month.When compared with the benchmark commercial solution called Kubecost, our solution outperformed by 33% on an average for CPU and 27% for memory across 12 microservices over a month of data.