End2End100-Phase 2

点击翻译按钮获取中文摘要

The processing of 100Gbit/s wireless data streams is a demanding task. Even today's parallel supercomputers struggle to deliver such tremendous throughput in wired networks end-to-end to the application level. In the case of wireless communication the situation gets worse, since multiple wireless channels with varying error rates need to be utilised in parallel to achieve the desired data rate. Therefore, the project End2End100 has focused on parallel MAC/data link protocols as well as a parallel NIC platform for ultra high-speed wireless communication in the first phase of the SPP. We modelled protocol processing tasks as soft real-time stream processing problems and mapped the corresponding processing stages to the cores of Tilera manycore CPUs. The time critical and computing intensive FEC processing tasks were off-laid to FPGAs. We designed and implemented flexible FEC schemes based on Reed-Solomon codes, that can potentially be adapted to the error rate of the wireless channel. Since no wireless communication platform was available in the first phase, we demonstrated the feasibility of our approach with a wired setup via 8x10Gbit Ethernet links. With this experimental setup we were already able to achieve 76Gbit/s out of the theoretically possible 80Gbit/s.In this follow-up project we will extend our concept with real wireless baseband processing and a PHY-layer. Thus, we plan to integrate at least the results of the Real100G.COM and Real100G.RF projects for a real wireless demonstrator (probably including suitable other projects). The data rate of 100 Gbps wireless will be achieved by combining several (individually error prone and unstable) wireless links. This poses two fundamental research objectives:1.) How can we achieve the highest possible throughput for a single unstable wireless link with a varying bit error rate?2.) How can we manage several wireless links with different qualities to achieve the highest possible throughput?We will solve this two-dimensional optimisation problem through self-adaptive efficient error correction schemes for the single link optimisation and self-adaptive multilink management by means of dynamic stream processing. An additional cross cutting objective is the reduction of the energy consumption with respect to specific protocol tasks, as well as the whole system. We will examine energy efficient processing strategies and exploit synergy effects between the higher level data link protocol processing, lower level data link processing, e.g., FEC calculation, baseband and the PHY-Layer. The continuation of our work requires funding for two full-time PhD-students, two part-time master students and travel funds for project meetings and conferences for three years. Additional hardware would ease our work significantly.