Conntrack offload: why and how
summary
Explains a concrete path to hardware-accelerated Conntrack by offloading the flow table (one per zone) to NICs, with control via the TC action CT; the pieces actually built are kernel support in kernel 5.7 and RHEL 8.3 tech preview with the Mellanox 5 driver as the only supported offloader so far. New connections still require software to create entries (offload is incompatible with skip-software), only established flows get flooded to hardware, and stats/aging remain managed by software.
AI-written summary · about
chapters (4)
related talks
21mAccelTCP: Accelerating Network Applications with Stateful TCP Offloading
21mRemote TCP Connection Offload and Applications
17mRearchitecting the TCP Stack for I/O-Offloaded Content Delivery
36mOffloading Network Traffic Classification to Hardware
27mHigh speed load balancing from the Linux Kernel
23mACK to the Future: TCP in 2025
14mFlexTOE: Flexible TCP Offload with Fine-Grained Parallelism
34mDeterministic Networking for Real-Time Systems (Using TSN)
