Collective Autoscaling for Cloud Microservices

Abstract: As cloud applications shift from monoliths to loosely coupled microservices, application developers must decide how many compute resources (e.g., number of replicated containers) to assign to each microservice within an application. This decision affects both (1) the dollar cost to the application developer and (2) the end-to-end latency perceived by the application user. Today, individual microservices are autoscaled independently by adding VMs whenever per-microservice CPU or memory utilization crosses a configurable threshold. However, an application user's end-to-end latency consists of time spent on multiple microservices and each microservice might need a different number of VMs to achieve an overall end-to-end latency. We present COLA, an autoscaler for microservice-based applications, which collectively allocates VMs to microservices with a global goal of minimizing dollar cost while keeping end-to-end application latency under a given target. Using 5 open-source applications, we compared COLA to several utilization and machine learning based autoscalers. We evaluate COLA across different compute settings on Google Kubernetes Engine (GKE) in which users manage compute resources, GKE standard, and a new mode of operation in which the cloud provider manages compute infrastructure, GKE Autopilot. COLA meets a desired median or tail latency target on 53 of 63 workloads where it provides a cost reduction of 19.3%, on average, over the next cheapest autoscaler. COLA is the most cost effective autoscaling policy for 48 of these 53 workloads. The cost savings from managing a cluster with COLA result in COLA paying for its training cost in a few days. On smaller applications, for which we can exhaustively search microservice configurations, we find that COLA is optimal for 90% of cases and near optimal otherwise.