Next Generation Telecommunications Computing Architecture (NexTCA)
Tai-Hua Hsiao, Chilung Wang, Peng-Kai Hsu · 2018
Many telecom services are deployed on proprietary and expensive equipment. Telecom operators must constantly invest in computing power and I/O capabilities to provide more services. However, it is not allowed to share resources among these dedicated systems, which leads to considerable waste of money and resources. A general-purpose platform should take care of virtualization, performance resource sharing and cost efficiency as a whole. This paper proposes NexTCA that embodies disaggregated rack architecture as well as network and storage MR-IOV (multi-root I/O virtualization) system based on PCIe technology by leveraging existing commodity hardware. Additionally, we develop an Ethernet over PCIe channel which fully uses PCIe link capability while providing transparent features to upper-layer applications. At last, OPNFV YardStick and FIO are deployed to verify the functionality and performance of NexTCA in comparison with a modern Ethernet-based system. The experiment results show that NexTCA remarkably outperforms a modern Ethernet-based system in both throughput and latency; even, PCIe is a superior technology for intra-rack and inter-rack communication. Reference from vendors and resellers also points out that the construction cost of PCIe is lower than that of 10Gbps Ethernet.