Continuous Packet Offloading with Update Rule
summary
New rte_flow_actions_update() API provides an atomic replacement of a rule’s action list so no packet is untagged or mis‑steered during updates. The PMD allocates new resources (counters, timers), stages a new action list off‑table, performs an atomic pointer swap in hardware, then frees the old resources. Implemented in Mellanox monolithic driver and available since 23.07, includes async/Q version; other PMDs are encouraged to implement the same pattern.
AI-written summary · about
chapters (7)
- 0:00Introduction to flow API
- 1:05Generic flow API overview
- 2:31Challenges with rule updates
- 4:59Priority-based workaround
- 7:06RT flow update API solution
- 9:38Implementation in Mellanox driver
- 10:41Q&A session
related talks
31mFlow-based Packet Processing Framework on DPDK and VPP
19mConntrack offload: why and how
22mLIVE Q&A #3
17mRe architecting Congestion Management in Lossless Ethernet
34mCoupling Eventdev Usage with Traffic Metering & Policing (QoS)
15mEnabling In-situ Programmability in Network Data Plane: From Architecture to Language
28mSafe sharing of the network with eBPF
16mPacket Order Matters! Improving Application Performance by Deliberately Delaying Packets