Building Scalable, High Performance Cluster and Grid Networks: The Role of Ethernet

WhitePAPER
Building Scalable, High Performance Cluster and Grid Networks:The Role of Ethernet
What Is the Grid? of computationally intense analysis needed for productresearch and development, and allows scientists to solveA Grid computer is a hardware and software system "grand challenge" problems that were too large for that integrates a collection of distributed system conventional supercomputers.components (e.g., computer systems, storage, etc.) In its current stage of evolution, most applications of themaking them appear to the user as a single, large Grid fall into the HPC classification. This is due to the"virtualized" computing system. The basic "single fact that Grid computing arose out of the need for moresystem" concept may be applied to the construction cost-effective HPC solutions to address critical problemsof centralized "cluster" computers (multiprocessor in science and engineering. The initial adoption of thesystems consisting of colocated computers and storage) Grid by commercial enterprises has continued to focusor a networked grid of geographically dispersed on HPC because of the high return on investment andcomputers, instrumentation, or other resources. competitive advantage realized by solving compute-With Grid computing, computer systems and other intensive problems that were previously insolvable in aresources aren't necessarily constrained to be dedicated reasonable period of time or cost. The HPC Grid hasto individual users or applications, but can be made successfully addressed a wide range of computationalavailable for dynamic pooling or sharing to meet the problems in the following fields:changing needs of the organization. By using the . Climate/weather/ocean modeling and simulationInternet, grid-based resource sharing and collaborative . Computational chemistry and materials scienceproblem solving can be extended to multi-institutional . Environmental quality modeling and simulation"virtual organizations." . Military forces modeling and simulationThe name "Grid computing" was chosen because the . Internet search enginesconcept is analogous to the electrical power grid, which . Pharmaceutical researchsatisfies customer demand for power by transparently . Simulation and verification of electronic and drawing on a distributed grid of power-generating mechanical designresources shared by all users. . Seismic processing and interpretation. Signal/image processing. Modeling of financial portfolios and marketsBenefits of the Grid Data Federation and Collaboration: The Grid alsoallows a federated approach to data integration whereBecause it is evolving as a very flexible, perhaps even data from different sources (relational databases, files, orall-encompassing, distributed system technology, Grid application data) can be consolidated in a single datacomputing offers a number of potential uses and benefits service that hides the complexities of data location, that can be broadly categorized in the following way: local ownership, and infrastructure from the consumingHigh Performance Computing (HPC): For applica- application. With data federation, the data remains intions that lend themselves to parallel computing, the place at its source, with no disruption of local users,Grid offers the possibility of executing computationally- applications, or data management policies. Integration intensive applications on networked computer arrays of data from multiple sources and locations facilitates aconsisting of numerous commodity or specialized wide range of integrated applications, including corpo-systems. Computational Grids can offer significant rate performance dashboards, marketing analysis tools,advantages in both price/performance and in maximum customer service applications, and data mining applica-performance over more conventional types of super- tions. Because the data resource is accessed as a servicecomputers. As a result, the Grid makes HPC accessible on the network, minimal modifications of existing data-to more enterprises, accelerates the availability of results publishing or data-consuming applications is required.
© 2 0 0 6 F O R C E 1 0 N E T W O R K S , I N C . [ P A G E 1 O F 1 5 ]Resource Allocation and Optimization: Most desktop (ISVs). While most of this software is proprietary, there issystems and servers are idle more than 80% of the time also a wide variety of open source middleware available.because computers have traditio... [download for more]