TCP offload is a dumb idea whose time has come

Abstract

Network interface implementors have repeatedly attempted to offload TCP processing from the host CPU. These efforts met with little success, because they were based on faulty premises. TCP offload per se is neither of much overall benefit nor free from significant costs and risks. But TCP offload in the service of very specific goals might actually be useful. In the context of the replacement of storage-specific interconnect via commoditized network hardware, TCP offload (and more generally, offloading the transport protocol) appropriately solves an important problem.