Cloud-Based High Performance Computing Clusters

2.5 Cloud-Based High Performance Computing Clusters

In the development history of information science from the last half a century, High Performance Computing (HPC) has always been a leading technology at the time. It has become a major tool for future innovations of both theoretical and research science. As new cross-disciplines combining traditional subjects and HPC emerge

2 Cloud Computing Technologies and Applications 31 in the areas of computational chemistry, computational physics and bioinformatics,

computing technology need to take a leap forward as well to meet the demands of these new research topics.

With the current financial crisis, how to provide higher computing performance with less resource input has become a big challenge for the HPC centers. In the construction of a new generation of computing center with high performance, we should not only pay attention to the choice of software and hardware, but also take fully account of the center operation, utilization efficiency, technological innova- tion cooperation and other factors. The rationality of the general framework and the effectiveness of resource management should also be fully considered. Only by doing these can the center gain long-term high-performance capacity in computing research and supply.

In other words, the new generation of high-performance computing center does not only provide traditional high-performance computing, nor it is only a high-performance equipment solution. The management of resources, users and vir- tualization, the dynamic resource generation and recycling should also be taken into account. In this way, the high-performance computing based on cloud computing technology is born.

The cloud computing-based high-performance computing center aims to solve the following problems:

• High-performance computing platform generated dynamically • Virtualized computing resources • High-performance computer management technology combined with tradi-

tion ones • High-performance computing platform generated dynamically

In traditional high-performance computing environment, physical equipments are configured to meet the demands of customers; for example, Beowulf Linux and WCCS Architecture are chosen to satisfy customers’ requirements on computing resources. All of the operation systems and parallel environment are set beforehand, and cluster management software is used to manage the computing environment. However, as high-performance computing develops, there are more and more end users and application software; thus, the requirements on the computing platform become more diverse. Different end users and application software may require dif- ferent operation systems and parallel environment. High-performance computing requires a new way of resource supply, in which the platform should be dynami- cally generated according to the needs of every end user and application software; the platform can be open, including Linux, Windows or UNIX.

• Virtualized computing resources Since few virtualized architecture are used in traditional high-performance com-

puting, this kind of platform cannot manage virtualized resources. However, as high-performance computing develops, in many cases we need to attain more

32 J. Zhu virtualized resources through virtualization, for example, the development and

debugging of parallel software, and the support for more customer application etc. In the cloud computing-based high-performance computing center, the virtual- ization of physical resources can be realized through the Cloud platform; moreover, virtualized resources can be used to establish high-performance computing platform and generate high-performance computing environment whose scale is larger than that of the actual physical resource so as to meet the requirements of customers.

• Combination with traditional management technology The cloud computing-based high-performance computing platform can not only

manage computers though the virtualization and dynamic generation technology, but also work together with traditional cluster and operation management soft- ware in enabling users to manage the virtualized high-performance computers in

a traditional way, and submit their own works.

A new IT resources provision model can be attained by the adoption of cloud computing infrastructure and high-performance computing center construction. In this model, the computing center can automatically manage and dynamically dis- tribute, deploy, configure, reconfigure and recycle the resources. The automatic installation of software and application can be realized, too. By use of the model, the high-performance computing resources can be distributed efficiently and dynam- ically. When the project is finished, the computing center can automatically recycle the resources to make full use of the computing power. Taking advantage of cloud computing, the high-performance computing center can not only provide high calcu- lating power for scientific research institutions, but also expand the service content of computing center. In other words, it can serve as a data center to support other applications and promote higher utilization efficiency of entire resources.