ADASCALE framework cuts microservice latency by 1.93x on Kubernetes

ADASCALE, developed by Ming Chen, Muhammed Tawfiqul Islam, Maria Rodriguez Read, and Rajkumar Buyya at the University of Melbourne, addresses the complex challenge of scaling and placing microservices in cloud-edge environments. These environments suffer from heterogeneous nodes, time-varying inter-node delays, non-stationary traffic, and shifting mixes of root request operations (e.g., compose-post vs. read-home-timeline) that exercise different call graphs. Existing autoscalers and network-aware schedulers handle only subsets of these dynamics, leading to compute bottlenecks or inflated cross-node latency and SLO violations. ADASCALE implements a Monitor–Analyzer–Planner–Executor (MAPE) loop that extracts per-edge and per-service demand from distributed traces and service-mesh metrics. It identifies the most critical root operations under mixed workloads, computes SLO-aware replica targets, and places replicas to minimize a demand-weighted latency objective given the current inter-node latency matrix. To react quickly to networking perturbations, ADASCALE triggers a reactive placement loop, while a steady-state autoscaling loop handles demand shifts.

Evaluated on a cloud-edge Kubernetes cluster using the DeathStarBench Social Network application with three root operations under varying load and workload mixes, ADASCALE consistently meets SLO targets and improves both latency and throughput. Compared to NetMARKS_Scale, ADASCALE achieves up to 1.56x, 1.93x, and 1.34x lower average response time for compose-post, read-home-timeline, and read-user-timeline respectively, and up to 2.16x, 1.32x, and 1.36x higher throughput respectively. The framework is particularly strong under high traffic and mixed workloads, where traditional schedulers fail. This work provides a practical solution for operators of microservice-based applications deployed across edge and cloud nodes, offering both reactive and steady-state mechanisms to handle real-world dynamics.