Netflix is utilizing eBPF to monitor Linux kernel performance and detect noisy neighbors efficiently, providing continuous insights with minimal performance overhead.
Using eBPF allows us to instrument the kernel and gather real-time insights about process scheduling and latency, overcoming traditional monitoring limitations.
We measure run queue latency for each container, essential for ensuring our workloads meet low latency requirements, and utilize three eBPF hooks to gather necessary data.
The collected data from instrumenting the kernel is processed through a Go program, which efficiently emits metrics to our Atlas backend.
[
Collection
]
[
|
...
]