Datacenter Design for Future Cloud Radio Access Network.

Cloud radio access network;Datacenter design;GPU;Resource management;Computer Science;Engineering;Computer Science and Engineering

Zheng, QiTang, Lingjia ;

University of Michigan

关键词: Cloud radio access network;    Datacenter design;    GPU;    Resource management;    Computer Science;    Engineering;    Computer Science and Engineering;   

Others  :  https://deepblue.lib.umich.edu/bitstream/handle/2027.42/120825/qizheng_1.pdf?sequence=1&isAllowed=y

瑞士|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Cloud radio access network (C-RAN), an emerging cloud service that combines the traditional radio access network (RAN) with cloud computing technology, has been proposed as a solution to handle the growing energy consumption and cost of the traditional RAN. Through aggregating baseband units (BBUs) in a centralized cloud datacenter, C-RAN reduces energy and cost, and improves wireless throughput and quality of service. However, designing a datacenter for C-RAN has not yet been studied. In this dissertation, I investigate how a datacenter for C-RAN BBUs should be built on commodity servers.I first design WiBench, an open-source benchmark suite containing the key signal processing kernels of many mainstream wireless protocols, and study its characteristics. The characterization study shows that there is abundant data level parallelism (DLP) and thread level parallelism (TLP). Based on this result, I then develop high performance software implementations of C-RAN BBU kernels in C++ and CUDA for both CPUs and GPUs. In addition, I generalize the GPU parallelization techniques of the Turbo decoder to the trellis algorithms, an important family of algorithms that are widely used in data compression and channel coding.Then I evaluate the performance of commodity CPU servers and GPU servers. The study shows that the datacenter with GPU servers can meet the LTE standard throughput with 4× to 16× fewer machines than with CPU servers. A further energy and cost analysis show that GPU servers can save on average 13× more energy and 6× more cost. Thus, I propose the C-RAN datacenter be built using GPUs as a server platform.Next I study resource management techniques to handle the temporal and spatial traffic imbalance in a C-RAN datacenter. I propose a ;;hill-climbing” power management that combines powering-off GPUs and DVFS to match the temporal C-RAN traffic pattern. Under a practical traffic model, this technique saves 40% of the BBU energy in a GPU-based C-RAN datacenter. For spatial traffic imbalance, I propose three workload distribution techniques to improve load balance and throughput. Among all three techniques, pipelining packets has the most throughput improvement at 10% and 16% for balanced and unbalanced loads, respectively.