Runtime Resource Management on Serverless Computing Architectures

Abstract

Function as a service or FaaS represents the next frontier in the evolution of cloud computing being an emerging paradigm that removes the burden of configuration and management issues from the developer’s perspective. However, this relatively new technology, like any other, surely comes with its caveats. For starters, the whole well known monolithic approach has to be replaced by a DAG of standalone, small, stateless, event driven components called functions. At the same time, at the cloud provider’s side, problems like availability, load balancing, scalability and others has to be resolved without ever knowing the functionality, behavior or resource requirements of their tenants code. In this context, vendors offer certain billing plans concerning the available resources (CPU, memory & cold storage size etc) of the containers/sandboxes that functions run on. Unfortunately, these containers have to coexist with others in a runtime of a host with finite shared resources. Thus, with the latter passive resource allocation technique there’s no guarantee of a well defined quality of service or QoS in regards to functions’ and function sets’ latency. Various efforts have been made towards the holy grail of QoS, but they either lack in compatibility with existing serverless frameworks, or they are limited in specific applications. In this thesis, we explore Sequence Clock, a latency targeting tool that actively monitors serverless invocations in a cluster and offers execution of sequential chain of functions, also known as pipelines or sequences, while achieving the targeted time latency. It was developed in Go, wrapped as a helm chart (a packaging format for Kubernetes) and focuses on OpenWhisk deployments on top of Kubernetes’ clusters. Two regulation methods were utilized, with one of them achieving up to a 82% decrease in the severity of time violations and in some cases even eliminating them completely.