Research And Design On Similar Socket Interface Under The DPDK Structure

With the continuous improvement of the Internet, the applications in the network have not only limited to the simple application of a small flow of data communication. A large amount of concurrency can be seen everywhere. Increasingly rich applications make the server of service provider, network equipment face serious challenges, it requires that the servers that run these applications and network equipment provide higher processing performance, greater scalability and reliability, and save more cost.In order to make the server and network equipment provide higher processing performance, current common solution is to increase the number of cores in the server, enhance parallel processing capabilities of the servers. But some experiments show that with the increasing of the number of system cores, data processing capability of the SMP system does not grow linearly with the increase from the number of cores, when the number of CPU added to a certain number, the network processing performance of the server only has minor improvements. So in this case, increase the server’s configuration blindly, to improve server performance obviously has not lead to a considerable increase in performance.With careful analysis it is not difficult to find that the reason of this problem is that the data processing performance bottlenecks of the traditional core protocol stack have been occured, it will increase with the increasing number of CPU cores, so the user only make the optimization of data transmission of application layer,it will not bring significant results.In the second chapter, through research and analysis of the design architecture of Linux kernel TCPIP protocol stack, we point out the bottlenecks and flaws of the traditional core protocol stack. Then we explore and analyze the Intel Data Plane Development Tools(DPDK), through the optimization of the drive, memory optimized, thread scheduling, and lock-free queue, we indicate that the basic principles of data plane acceleration. Here, we will do a basic forwarding test using ovs and dpdk-ovs, to demonstrate the stronger packet data processing performance of DPDK.The third chapter is the main part of this thesis, it is the key work of this paper. We will complete three part of the design. First, we realize a lightweight TCPIP protocol stack in the user mode based on the data plane API DPDK provided, to make up for the deficiency that there is not stack in DPDK. It implements basic protocols such as IP, UDP, TCP, ARP, and ICMP, to provide support for the upper application; Secondly we provide Socket API for the upper application as static libraries. It will communicate with the protocol stack processe by lock-free queue. Third, we will transplante TFTP server based on UDP and TCP-based open-source HTTP server.Finally, we will do a detail test for the function and performance of TFTP server and HTTP server in the paper. The test result shows that in the same configuration of server the TCPIP protocol stack based on the DPDK will provide stronger data processing performance than the Linux kernel protocol stack.