* BPF TCP-CC and more There is a constant need to change the network protocol (and its underlying algorithm) to make improvements. Protocol ossification is one of the well-understood production hurdles. Even within a well-self-contained network environment like data-center, it is challenging to rollout out this kind of changes while a long tail of kernel version running. Also, another question is how to enable different expertise (e.g. TCP Congestion Control expert) to implement kernel changes easier and safer. The above essentially limited the speed of implementing and rolling out network-protocol related kernel changes. BPF has become a natural solution for these. This talk will first discuss about BPF STRUCT_OPS which was first introduced to allow writing TCP CC in BPF. What has been done and which parts still need some works. Another discussion of this topic is about what other parts of the network protocol could be made extensible by BPF. For example, one idea is to make TCP header options extensible by BPF.
