RAYGO: Reserve As You GO

The capability to predict the precise resource requirements of a microservice-based application is a very important problem for cloud services. In fact, the allocation of abundant resources guarantees an excellent quality of experience (QoE) for the hosted services, but it can translate into unnecessary costs for the cloud customer due to the reserved (but unused) resources. On the other side, poor resource provisioning may turn out in scarce performance when experiencing an unexpected peak of demand. This paper proposes RAYGO, a novel approach for dynamic resource provisioning to microservices in Kubernetes that (i) reliefs the customers from the definition of appropriate execution boundaries, (ii) ensures the right amount of resources at any time, according to the past and the predicted usage, and (iii) operates at the application level, acknowledging the dependency between multiple correlated microservices.