Improving task scheduling in microservice environments by considering intra-job dependencies

Microservices in modern-day servers usually operate within container-based environments, which run on dedicated computing nodes. Even though these nodes are responsible for running containers belonging to a microservice, the scheduler of the OS running on a node is unaware of the intra-job dependencies in containers that arise to serve a particular request, instead of treating them like independent processes. Our solution takes advantage of request propagation data obtained from tracing tools and can adjust resource allocation based on the throttle time of a container. This allows us to efficiently allocate CPU resources to containers across services and automatically adjust scheduling parameters to provide better response times.