Quantifying Autoscaler Vulnerabilities: An Empirical Study of Resource Misallocation Induced by Cloud Infrastructure Faults

TL;DR

This study quantifies how common cloud infrastructure faults distort autoscaling decisions and drive resource misallocation. Using controlled simulations across four fault types, two autoscaling strategies, multiple instance families, and two SLO thresholds, it reveals that disk-related faults cause the largest cost overhead under horizontal scaling, while routing faults tend to push toward underprovisioning; horizontal autoscaling is more sensitive to transient anomalies near decision thresholds. The work provides actionable mitigation strategies, such as multi-metric validation and latency- and I/O-aware triggers, and demonstrates that SLO choices significantly influence the magnitude of misallocation and cost. These findings offer practical guidance for designing fault-tolerant autoscalers that separate genuine workload variation from failure artifacts, ultimately reducing cost risk while maintaining service reliability.

Abstract

Resource autoscaling mechanisms in cloud environments depend on accurate performance metrics to make optimal provisioning decisions. When infrastructure faults including hardware malfunctions, network disruptions, and software anomalies corrupt these metrics, autoscalers may systematically over- or under-provision resources, resulting in elevated operational expenses or degraded service reliability. This paper conducts controlled simulation experiments to measure how four prevalent fault categories affect both vertical and horizontal autoscaling behaviors across multiple instance configurations and service level objective (SLO) thresholds. Experimental findings demonstrate that storage-related faults generate the largest cost overhead, adding up to $258 monthly under horizontal scaling policies, whereas routing anomalies consistently bias autoscalers toward insufficient resource allocation. The sensitivity to fault-induced metric distortions differs markedly between scaling strategies: horizontal autoscaling exhibits greater susceptibility to transient anomalies, particularly near threshold boundaries. These empirically-grounded insights offer actionable recommendations for designing fault-tolerant autoscaling policies that distinguish genuine workload fluctuations from failure artifacts.

Figures (4)

• Figure 1: Architecture of the experimental environment, which consists of a server group that includes the experimental group and control group, a client group, and the simulation tools.
• Figure 2: Comparison of the gap in the optimal specification between normal and failure state by instance size for vertical autoscaling with (a) SLO 85% and (b) SLO 50%. Failures include SYN flooding attack (SYN), UDP flooding attack (UDP), volumetric attack (Vol), router failure (Rtr), disk failure (Disk), and software problem (App).
• Figure 3: Comparison of the gap in the optimal replicas count between normal and failure state by instance size for horizontal autoscaling with (a) SLO 85% and (b) SLO 50%. Failures include SYN flooding attack (SYN), UDP flooding attack (UDP), volumetric attack (Vol), router failure (Rtr), disk failure (Disk), and software problem (App).
• Figure 4: Comparison of the gap in monthly optimal cost between the normal and failure states, by failure type in (a) vertical autoscaling and (b) horizontal autoscaling, comparison of the ratio of resource insufficient risk caused by the failure and expense, by instance types in (c) SLO 85% and (d) SLO 50% vertical autoscaling for total failures.