Research On Performance Evaluation And Optimization Technology Of Multi-core Processing System Based On DPDK

Data Plane Development Kit(DPDK)provides an effective solution for fast packet processing of general-purpose multi-core processor platforms.Based on the DPDK research on general-purpose multi-core processor platforms,CPU cache can be utilized to improve packet forwarding and processing performance,as well as network function virtualization performance.At present,the problem to be solved include: first,how to carry out the analysis and modeling of the DPDK-based general-purpose processing system;second,the performance optimization based on fully combining the hardware characteristics.Aiming at the performance modeling analysis and performance improvement of DPDK general data packet processing system,the main research contents are as follows:(1)A multi-core processor platform performance analysis model based on DPDK affinity is proposed.In view of the fact that the existing performance analysis model does not consider the characteristics of DPDK processing system node distribution and affinity binding,which thus deviates from the actual system,a multi-core processing system queuing model Multi-core System Queuing Model based on DPDK Affinity Setting(MQDAS)is proposed.The model considers the fixed binding between the network interface card queue,the processing thread and the processor logic core.The experimental results show that the proposed model can more accurately describe the actual performance of multi-core processor systems based on DPDK optimization technology.In addition the application analysis model can determine the number of processing cores allocated to reduce packet loss rate and thus improve processing performance.(2)A task scheduling optimization algorithm for DPDK Pipeline processing model is proposed.The existing multi-core task scheduling algorithm only considers the success rate of scheduling without the hardware characteristics of the system,resulting in unbalanced processing core utilization and affecting the processing performance of the system.Therefore this thesis proposes a task scheduling optimization algorithm Task Scheduling Optimization Algorithm for DPDK Pipeline Processing Model(TSOA-DP).When the task is assigned to the processing core,the hardware distribution of the processing core is considered.The experimental results show that the task scheduling algorithm proposed in this thesis improves the scheduling success rate by 3%-4% while the balance of processing core utilization rate is also improved.(3)Combined with the performance model and task scheduling optimization algorithm proposed above,a DPDK-based rapid processing prototype system for user space messages is designed and implemented.The system consists of a high-speed message generation sub-module,a queue pre-allocation sub-module,a packet fast processing sub-module,and a task scheduling sub-module.The experimental results show that the data transmission rate of the prototype system reaches more than 10 Mpps and the processing rate of the receiver and forwarding reaches more than 7Mpps.The application of the evaluation model and the optimized task scheduling algorithm improves the processing performance of the multi-core processing system implemented in the DPDK user space.