Network Packet Processing with P4 (Programming Protocol-Independent Packet Processors) in CPU, GPU and FPGAs using OpenCL

点击翻译按钮获取中文摘要

The purpose of Network functions virtualization (NFV) is to virtualize the network services that are now being offered by dedicated hardware. NFV will decrease the amount of dedicated hardware that's needed to launch and operate network services. In other words, NFV is proposed to decouple network functions from dedicated hardware devices and allow network services that are now being carried out by routers, firewalls, load balancers and other dedicated hardware devices to be hosted on virtual machines (VMs). Consequently, the network administrators will no longer need to purchase dedicated hardware devices in order to build a service chain. As a result, server-functionality and capacity is thus dictated by the available generic hardware and the software-enforced quota of that hardware granted to an NFV; there will be no need for network administrators to over-provision their data centres with specialized single-use hardware, which will reduce both capital expenses and operating expenses. Intel CPUs are the currently available "generic hardware" being used for NFV applications; Intel CPUs are not designed for highly parallel tasks like packet processing. Bleeding edge technology investments by Intel suggest that future "generic hardware" for NFVs will have FPGA (Field Programmable Gate Arrays) or GPU (Graphics Processing Units) based co-processors for offloading highly parallel tasks. An important consequence of the availability of this class of hardware is that designers could configure functionality after manufacture and even after deployment should this be desired by the customer. Emerging technologies for heterogeneous computing (OpenCL) will be investigated as potential technology investments to target current generation hardware and next generation hardware with the same code-base.