الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 240 |  |  |
| | | from | 240 | | |
Abstract
High Performance Computing (HPC) is the cornerstone for any scientific and industrial breakthroughs. HPC clusters address the demand of highly reliable, dedicated, and real-time processing IT infrastructure to process, analyze, and evaluate huge amounts of data in compute and data intensive applications. Most HPC applications recently started utilizing the cloud resources, called (HPC cloud) where building a parallel Virtual Cluster (VC) to move and run HPC applications on the cloud. In this approach, Infrastructure as a Service (IaaS) cloud solution is adopted for HPC users. Some compute-intensive applications target cloud features such as centralized processing and management, as well as dynamic scalability and resources utilization. Both HPC and cloud computing have in common the advantage of computationally powerful servers and the capability of parallel processing for compute-intensive applications in a reasonable time. The wireless communication is one of the sectors that utilizes cloud computing and HPC functionalities to allow the fifth generation of cellular communication (5G) to be introduced.
5G systems are considered the future of telecommunications and data processing. Due to rapid proliferation of mobile social applications, data traffic from mobile devices is continuously and sharply increased in the foreseeable future. Besides, the main goals of fifth generation (5G) systems are to handle the enormous network capacity and consider new 5G applications requirements such as high data rate (10 GB), low latency (1 ms), high reliability, enhanced spectral efficiency (fixed data rate per bandwidth), and energy efficiency (EE). Therefore, one of the major challenges of 5G network management is to have a central, efficient, and parallel multi-hop routing protocol for 5G networks to easily manage and speed up the routing decisions and data transfer among dense communication nodes (small cells) with taking into account different 5G applications requirements. Since the best route selection process in ultra-dense network (UDN) involves intensive computation and time-consuming tasks.
The main thesis contribution is to propose a parallel multi-hop routing protocol that selects the optimal route while achieving the 5G needs such as spectral efficiency target and using the minimum transmission power between any two communication nodes in the selected route. We use Message Passing Interface (MPI) as the parallel programming model. The parallel protocol aims to speed up routing decisions in 5G networks, using HPC cluster and virtual cluster on cloud in 5G core networks. Herein, we study the efficiency of utilizing cloud computing and HPC platforms to manage and speed up the parallel routing protocol for different communication network sizes and MPI distributing scenarios. Recommendations are set for utilizing both platforms to adopt the parallel protocol.
Our results indicate that the scalability of HPC outperforms that of cloud computing in terms of the achieved routing speed-up. In particular, for a large network size with 2048 nodes, our HPC implementation achieves a routing speed-up of 37x, and is recorded using a few hardware requirements (one VM) for a medium network size of 1024 nodes. By comparing the results of our protocol for both platforms with the original serial algorithm, we conclude that there is a marked improvement in running bigger network sizes with less execution time. To set recommendations for the adoption of cloud computing and HPC platforms to be highly utilized in terms of speed up, scalability, and resources or infrastructure requirements in 5G deployment, a set of experiments are done using different application and system parameters (network size, number of cores and nodes). We conclude that our parallel protocol is a latency-sensitive application, and cloud virtualization adds more overhead. Accordingly, the cloud computing outperforms HPC that it could be deployed at the edge of medium to large networks to serve mobile users and real-time applications, yielding a peak speed up with a few hardware requirements (one VM). We also found that the location of VMs and the number of VCPUs per VM in the cloud server play a crucial role in determining the latency and virtualization overhead for different network sizes. Therefore, deploying virtual cloud clusters necessitates enhancing the cloud resource management to reduce network virtualization overhead, especially for our parallel 5G routing protocol. We conclude that our HPC cluster implementation surpasses that of a virtual cloud cluster in terms of routing speed-up for a large network size. Therefore, HPC is more suitable than the cloud computing in terms of higher linear parallel performance, scalability, and platform utilization at the core network of 5G.