Work

Multi-Signal SLA-Driven Autoscaling for Elastic Cloud-Native Workloads

Year: 2026

Type: conference-paper

Abstract: In modern cloud-native artificial intelligence (AI) assisted infrastructures, services require responsive autoscal ing mechanisms capable of maintaining service-level objectives (SLOs) under highly dynamic and bursty workload conditions. However, conventional autoscaling approaches often evaluate op erational signals independently, limiting their ability to capture the combined effects of service degradation, workload pressure and backlog accumulation during inference execution. In this paper, we present an autoscaling framework for cloud-native workloads driven by service-level agreement (SLA) require ments. The proposed approach enables autoscaling decisions through a composite infrastructure pressure metric that jointly considers service latency, workload intensity and system backlog conditions. The framework is implemented as a cloud-native external autoscaling service integrated with Kubernetes Event Driven Autoscaling (KEDA) and observability pipelines, enabling adaptive and lightweight orchestration within cloud-edge envi ronments. Experimental evaluation conducted on a Kubernetes testbed demonstrates improved resource efficiency and more stable scaling behavior compared with conventional autoscaling approaches.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.23008333

Authors: Selis, Dimitrios, Ramantas, Kostas, Luis Alonso, Verikoukis, Christos

Institutions: Industrial Systems Institute, Athena Research and Innovation Center In Information Communication & Knowledge Technologies, Iquadrat (Spain), General University Hospital of Patras

Language: English

Topic: Cloud Computing and Resource Management

Subfield: Information Systems

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7214677655

FWCI:

Cites:

Cited by:

Related to: 10