Heterogeneous Cloud-Computing Architecture for High-Performance, Security and Advance Networking Research
Abstract
Cloud computing allows companies to access computing and storage infrastructure on demand, thus eliminating acquisition and maintenance costs, while increasing flexibility and availability of their solutions. Current cloud infrastructure providers such as Amazon rely exclusively on General Purpose Processors (GPP) as computational engines underneath virtual machines deployed to service customers. Graphics Processing Units (GPUs) and Reconfigurable Processors, such as FPGAs, outperform general-purpose processors in many applications that require graphics, vector processing, and data streaming. Thus far, cloud-computing companies have not been able to provision heterogeneous computing resource to customers, despite their usage in some data centers such as Microsoft for dedicated search acceleration. The main reason is the lack of design and management tools. Customers, most whom do not have the require skills, must rewrite their code to utilize the specialized hardware. To overcome these limitation, the PI will integrate of the three primary computation paradigms (GPP, GPU, and Reconfiguration) into a cloud infrastructure to provide near-optimal computing architectures on-demand in an ÓInfrastructure as a ServiceÓ-solution, facilitated by intelligent methods that dynamically adapt resources to run-time computational and power consumption requirements. The resulting infrastructure will enable a wide range of research activities at the University of Arkansas and nationwide in the following areas, all of which are areas of interest for the DoD. ¥ high-performance computing, ¥ bid data analysis, ¥ security, and ¥ networking. To reach the goal of this research, the PI will design and build a heterogeneous cloud computing infrastructure with 25 nodes, each of which will consist of a multicore CPU, an FPGA, a GPU, memory, high speed intra and inter-node interconnect, and network storage. The entire cloud will be operated using the open source cloud management package Openstack, with significant parts of the integrated KVM virtual machine implemented into hardware for security and performance improvement. The proposed architecture will extend the capabilities of current cloud computing solutions and make them more attractive for interdisciplinary research in the fields of computer architecture, data analytics, image processing, security, and financial trading. To allow non-experts to leverage the proposed architecture diversity in cloud computing systems for their applications, the PI will provide methods and tools for migrating legacy software to the heterogeneous cloud nodes allocated to a user.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jun 10, 2016
- Source ID
- N000141612014
Entities
People
- Christophe Bobda
Organizations
- Office of Naval Research
- United States Navy
- University of Arkansas System