Exploiting DPDK in Containerized Environment with Unsupported Hardware

Abstract

Network virtualization is an attractive technique to deploy new network services in an agile and cost efficient way. However, since virtualization imposes additional performance overhead (e.g., disk input/output virtualization overhead), which results in higher latency for service deployment, new virtualization frameworks that accelerate the performance of virtualized network functions are becoming available. Among these frameworks, Intel has proposed a set of libraries to accelerate packet processing and to remove additional delays caused by context switching from kernel space to the user space in computing servers. In their current public releases, DPDK libraries are guaranteed to work only if specific requirements in terms of supported hardware and Network Interface Card (NIC) are satisfied. However, the supported physical NICs are usually found in high performance servers. Therefore, it will be more cost-effective (and useful for research activities) to be able to deploy DPDK using any physical NIC. In this study, we aim at demonstrating that DPDK experimental activities can be run also on unsupported hardware (i.e., hardware that is not originally supported by DPDK libraries). To demonstrate this, we propose various component stack and implement different testbed setups to exploit DPDK in our lab environment using generic servers. Results obtained show that, even on quite outdated and baseline equipment available in our lab environment, it has been possible to run DPDK and, using DPDK, we were able to reach the line rate assigning at least two CPU cores to DPDK application and using packet size greater than 256 Bytes.