The Compute and Performance Engineering teams at Netflix tackle performance issues in their multi-tenant environment by leveraging eBPF to monitor noisy neighbor problems. This blog post details how they use eBPF to continuously instrument the Linux scheduler to catch run queue latency issues efficiently. By accessing kernel data structures through eBPF, the teams can identify noisy neighbor issues, especially when system services compete with containers for CPU time. Additionally, the teams developed an open-source eBPF monitoring tool called bpftop to optimize their eBPF code and improve calculation accuracy. The findings demonstrate the value of low-overhead eBPF instrumentation for enhancing infrastructure observability and refining CPU isolation strategies.
https://netflixtechblog.com/noisy-neighbor-detection-with-ebpf-64b1f4b3bbdd