SSDK SDN Policy Offload Pipeline Overview

SDN Policy Offload Reference Pipeline User Guide (UG1668)

Document ID
UG1668
Release Date
2024-02-15
Revision
1.1 English

The SDN Policy Offload reference pipeline provides an example of how to use an AMD-developed, ready-to-deploy P4 library to build a high-performance, highly scalable data path application. The P4 library implements the routing table and ACL security lookup in P4-DMA, achieving much higher performance compared to performing the legacy approach where the routing table lookup and ACL lookup take place on CPU cores. By storing the routing tables and ACL table on the DDR memory, the solution can scale to millions, or tens of millions of entries without the high cost of HBM or TCAM used in ASIC designs. Connection tracking and flow aging are handled in the P4 pipeline, further offloading the CPU unlike legacy designs. With the routing table lookup and ACL lookup offloaded and accelerated by P4-DMAs and connection tracking offloaded by P4, the P4 library allows you to build a stateful services application with very high CPS or stateless services applications (not flow-aware or connection session aware) such as cost-effective router with large scale routing table and ACL. You can do various manipulations on the packet in the dataplane, but before the flow is installed, the packet must go through DMA for the route lookup and policy evaluation and with the option to even enforce all packets of a flow to go through route and policy evaluation.