A Counting-Based Approach to Scalable Micro-service Deployment

Deploying a cloud-based distributed application created from the composition of micro-services is a challenging problem. It mandates the resolution of a resource allocation problem accounting for resource utilization and network load. But it also imposes security requirements such as the selection of suitable technology stacks to protect the communication channels. Both sets of decisions are intimately related as hosting decisions affect the cost or feasibility of security measures under consideration. This paper revisits the problem and focuses on a scalable approach suitable to deploy large distributed applications. Specifically, it introduces a counting-based model to deliver solutions for hundreds of services within short computation times. The essence is to side-step some of the difficulties by focusing first and foremost on deciding how many services of each type need to be deployed at each location and postponing the instance connectivity problem to a post-optimization phase. Empirical results demonstrate the scope of the improvements and illustrate the performance to expect as a function of instance sizes.