Abstract
As different eBPF libraries keep emerging, developers are left with the hard task of choosing the right one. Until now, this choice has been based on functional requirements (e.g., programming language support, development workflow), while quantitative metrics have been left out of the equation. In this paper, we argue that efficiency metrics such as performance, resource usage, and data collection fidelity also need to be considered for making an informed decision. We show it through an experimental study comparing five popular libraries: bpftrace, BCC, libbpf, ebpf-go, and Aya. For each, we implement three representative eBPF-based tools and evaluate them under different storage I/O workloads. Our results show that each library has its own strengths and weaknesses, as their specific features lead to distinct trade-offs across the selected efficiency metrics. These results further motivate experimental studies to increase the community's understanding of the eBPF ecosystem.