DPDK-based implementation of application-tailored networks on end user nodes

Application-tailored networks are customized networks optimized for application requirements. They use custom protocol stacks and network virtualization to provide flexible and efficient communication. End user nodes run a framework called NENA to connect to such networks at runtime. The current NENA implementation runs on top of the operating system´s network stack and uses the Socket API. It allows to show the flexibility of application-tailored networks but limits NENA´s performance. Packet I/O frameworks like Intel´s Data Plane Development Kit (DPDK) offer direct access to network interfaces and high performance. So, such a packet I/O framework provides an opportunity to enhance NENA and investigate its performance. In this paper, we present a new implementation of NENA with Intel´s Data Plane Development Kit and evaluate its performance. Experiments show that our approach can achieve more than 9 times higher packet rates than a Socket-based implementation and is able to saturate a 10 Gbit/s link using two CPU cores.