As adoption of Kubernetes increases, organizations are now finding themselves deploying network overlays to integrate fleets of Kubernetes clusters. There are already a lot of choices for achieving that goal, but with the rise of extended Berkeley Packet Filter (eBPF) technology it will soon become possible to run networking software in-kernel in a way that makes it easier to scale networking services. There are several use cases for eBPF, including traffic control, creating network policy, connect-time load balancing and observability. Without eBPF, packets use the standard Linux networking path on their way to a final destination.

If a packet shows up at point A and you know that the packet needs to go to point B, you can optimize the network path in the Linux kernel by sending it straight to point B. With eBPF, you can leverage additional context to make these changes in the kernel so that packets bypass complex routing and simply arrive at their final destination.

This is especially relevant in a Kubernetes container environment, where, in addition to the host network stack, each container has its own mini network stack. When traffic comes in, it is usually routed to a container stack and must travel a complex path as it makes its way there from the host stack. This routing can be bypassed using eBPF. 

• The current state of container networking
• The rise of eBPF and its impact on container networking
• Things to consider when using eBPF