eBeeMetrics: An eBPF-based Library Framework for Feedback-free Observability of QoS Metrics

Researchers from academia have introduced eBeeMetrics, a novel open-source framework that addresses a fundamental challenge in system observability. Many critical system management runtimes (SMRs) for resource and power management rely on Quality-of-Service (QoS) metrics like tail latency and throughput as feedback. However, these metrics are notoriously difficult to observe directly; they aren't available via standard hardware performance counters and require complex, often invasive, application instrumentation to measure, creating significant overhead and integration complexity.

eBeeMetrics solves this by leveraging the extended Berkeley Packet Filter (eBPF) technology within the Linux kernel. The framework accurately derives application-level QoS metrics by observing only eBPF-accessible events, such as system calls. This approach eliminates the need to modify the application being monitored. The tool can act as a drop-in replacement to decouple SMRs from direct QoS feedback or supplement existing metrics to better identify server-side dynamics. According to the paper accepted to ISPASS 2026, eBeeMetrics demonstrates a strong correlation with real-world measured throughput and latency across various latency-sensitive workloads, validating its accuracy without the traditional instrumentation burden.