Metrics Visualization Using Profiling-Based Adaptive GC Policy for Serverless

Abstract

In serverless, applications operate under strict performance and resource-usage constraints, making memory efficiency a critical aspect of system design [1]. One approach to improving efficiency is to reduce the memory footprint through informed garbage-collection (GC) decisions [2]. In this work, we present the GC Hints system that enables dynamic, workload-aware GC triggering using GC Hints policy [3] and introduce a new monitoring and visualization framework in serverless workloads. The GC Hints system is implemented within the GraalVM Serial GC [4]. It collects detailed runtime statistics, GC counters, memory-usage trends, and per-request execution times. These metrics are captured at the beginning and end of each request, enabling detailed analysis of application behavior. The visualization framework shows heap occupancy and GC activity over time, allowing us to gain deeper insight into application execution patterns and to evaluate the effectiveness of the GC Hints policy in comparison to the default policy. The visualizations also serve as a diagnostic tool, highlighting outliers or unexpected program behavior. We evaluate the system using a suite of benchmarks [5] and microbenchmarks. Our results demonstrate that the policy improves memory-usage characteristics and contributes to more predictable performance across representative serverless workloads.