Network Function Decomposition and Offloading on Heterogeneous Networks With Programmable Data Planes

Programmable network hardware is emerging as a viable option for offloading and thus accelerating functions. However, the heterogeneous resources available in calls disaggregated deployment approach. switches, programmable Network Interface Cards (NICs), in-network compute nodes exposes different peculiar capabilities that can be maximally exploited only if functions are decomposed into multiple smaller This work presents framework automatic of The comprises an orchestrator capable deploying on software switches running containers. exploits optimization model choosing best decomposition according to traffic demands, topology, other constraints. An improved seamless heuristic also accounts necessity rerouting when need re-programmed. Furthermore, provides tool combine single P4 program via template pipeline deployed switch. Numerical results highlight advantages hardware. we show how have negligible effects allocation time accepted requests while guaranteeing put maintenance mode.