AIMS: Robustness Through Sensible Introspection

Publication
Proc. of the ACM SIGOPS European Workshop
Abstract

AIMS is an adaptive introspective management system designed to improve the robustness of complex distributed applications by enabling dynamic, fine-grained monitoring and adaptation. Rather than relying on static instrumentation or fixed monitoring strategies—which quickly become inadequate in heterogeneous, rapidly changing environments—AIMS allows probes to be installed, removed, reconfigured, and redeployed at runtime. It integrates customizable filters, predictive analytics, and both passive and active sensors to generate application-specific metrics and trigger context-appropriate adaptations. AIMS supports a hierarchical feedback model in which system resources are monitored and controlled at one level, while the effectiveness of those mechanisms is evaluated and adapted at a higher level. By enabling lightweight, dynamically adjustable introspection, AIMS reduces over- and under-instrumentation and supports more resilient decision-making in environments such as cluster-based Internet services, where resource conditions and workloads fluctuate unpredictably.