ABSTRACT
II
ABSTRACT
In recent years, the demand for individual users' network traffic has been increasing,
but the revenue of traditional mobile operators is not optimistic. The reason is that the
higher and higher network requirements of users have caused the cost of deploying new
equipment to be continuously increased in operations. Controlling the increase in the cost
of the underlying equipment to reduce costs has become a topic of common concern to
operators, and Network Function Virtualization (NFV) technology has emerged. As a
widely used technical method, it decouples the network function hardware held by
traditional operators and separates the network function from the proprietary hardware. It
is deployed on X86 general-purpose servers and depends on the software. Implement
corresponding virtual network functions to provide services outward. In this way, the
resources can be effectively scheduled according to real-time requirements and the need
for reducing equipment costs can be achieved.
However, in actual application deployment, there are still many problems in the use
of specific NFVs. For example, general-purpose servers are generally used to process
traditional IT services. Compared to proprietary hardware, they handle high-rate
forwarding services in NFVs. Facing severe performance bottlenecks, it is necessary to
analyze in a targeted manner how to increase the efficiency of NFV use in actual use. In
a single server, there are multiple processes in the entire forwarding process of a data
packet that can affect the use efficiency of network functions. Using different
virtualization methods will also have different effects on the deployment and migration
of subsequent network functions.
This thesis focuses on the performance requirements of NFV in actual deployment,
analyzes the actual deployment architecture, and analyzes the data forwarding plane and
virtualization plane that may cause performance bottlenecks. In the virtual forwarding
layer, three software switching technologies, DPDK-OVS, NetMap, and Click, which are
widely used in the industry, were selected for deployment. The technologies of DPDK
and OpenvSwitch were integrated, and the use of NetMap in the NFV scenario was
improved. The internal data structure, which builds the Click software router by
constructing the Click infrastructure chain, implements three data forwarding planes, and
compares the performance of the three data planes. In the selection of virtualization
评论0