Research On Congestion Control Method Based On Programmable Data Plane

In recent years,the rise of emerging network technologies such as the Internet of Things,blockchain,and cloud computing not only brings more convenience to people’s daily lives,but also puts forward higher requirements for the service capabilities of modern networks.The traditional network adopts a distributed architecture,and its data plane and control plane are tightly coupled,and each network element device executes the corresponding data packet forwarding logic according to the routing management policy pre-programmed by the manufacturer.This makes the entire network in a closed state,and it is difficult for the system to obtain the overall information of the entire network,which reduces the interoperability and scalability of network element equipment,increases the complexity and management difficulty of the network,and hinders the development of new network technology applications.The emergence of Software-Defined Network(SDN)and programmable data plane provides a new idea for the development of future network technology.SDN realizes centralized control of network state by decoupling the data plane and control plane.The programmable data plane technology represented by P4(Programming Protocolindependent Packet Processors)and POF(Protocol Oblivious Forwarding)endows the data plane with programming capabilities.The programming control of the switch realizes the corresponding functions.This greatly improves the flexibility and scalability of the network,and is conducive to the realization of functions such as traffic scheduling,congestion control,and security management at the system level.This topic adopts the related technologies of SDN and programmable data plane to carry out research on the problem of network congestion control.The main work is as follows:(1)Based on POF and RuleMap packet classifier,a high-performance packet classification scheme supporting multi-protocol is proposed.The scheme adopts the idea of flow table storage and hierarchical parallel search and matching based on field segmentation,divides the matching field into several subfields flexibly,and sets the corresponding matching unit to store the flow table information of the corresponding field.Each matching unit only matches field values in a specific range of the corresponding field,and generates an index table to improve the matching speed.Compared with the RuleMap packet classifier,this solution can effectively reduce the invalid search times and storage space occupation of the flow table,improve the search and matching efficiency of switches,reduce resource overhead,and help reduce queue delay,thereby alleviating network congestion.(2)Based on P4 and sliding window mechanism,a congestion control method suitable for programmable data plane is proposed.With the help of the metadata provided by the P4 switch,the controller can obtain the queue depth and connection status of each switching node in real time,as well as the transmission time,priority,data packet flow and other information of each service flow.The system completes the identification and processing mechanism of custom data packet headers by programming the switch,and realizes the congestion window control of different business flows.This solution comprehensively considers the network link status and service priority information,and dynamically adjusts the rate of each data stream in a differentiated manner,which can achieve fine-grained congestion control.