Traffic Footprint Characterization of Workloads Using BPF

Explore a 26-minute conference talk from the Linux Plumbers Conference that delves into characterizing traffic footprints of workloads using BPF (Berkeley Packet Filter). Learn how Aditi Ghag from VMware proposes using eBPF programs to detect and attribute network elephant flows to containers or VMs, aiming to improve scheduling and network performance of containerized workloads. Discover the challenges of network resource characterization for diverse application workloads, and understand how this approach can help separate mice flows from elephant flows to achieve low latency. Gain insights into using learned network footprints for dynamic allocation of Linux network resources, such as bandwidth and receive-side scaling queue mappings, all without modifying kernel code.

LPC2019 - Traffic footprint characterization of workloads using BPF