Researchers at eight U.S. universities are collaborating in the development of Archer, a community-based computing infrastructure for computer architecture research and education. Archer addresses, in an integrated manner, computer architecture researchers' broad need to access powerful computational resources and share simulation environments. The project achieves this objective by deploying a large-scale, flexible, and easy-to-use computing infrastructure which adds users' local computing resources to Archer-dedicated resources as users join the network, thereby increasing Archer's computing power as more users join. In addition, the Archer project comprehends a repository where users can share tools and simulation environments they've created, adding another incentive for users to join the Archer community, facilitating collaboration among research groups, and advancing the field of computer architecture.
The core hardware of Archer consists of computer clusters to be deployed at Cornell University, Florida State U., Northeastern U., Northwestern U., U. FLorida, U. Minnesota, and U. Texas at Austin. The hardware infrastructure provides hundreds of processors to support the demanding simulation workloads required by computer architecture research in areas including soft-error modeling, statistically-based design exploration, and chip-multiprocessor micro-architectures. The key software technologies enabling community resources to be aggregated in a secure, scalable and seamless manner are resource virtualization and robust batch job scheduling, which include the IPOP virtual network from U. Florida and the Condor middleware from U. Wisconsin.
The Archer community resource will enable advances in computer architecture by allowing researchers, particularly those in institutions not equipped with local resources for high-throughput computing, to run large-scale simulation experiments. Its technology provides a new way to swiftly create ad-hoc shared "Grid" computing pools within or across institutions, which will pave the way to the creation of similar cyberinfrastructures for other communities.
This project has resulted in novel approaches to deliver computational capabilities to scientists, researchers and students in a manner that greatly simplifies the deployment and use of computing resources. The typical platform for "high throughput computing" is a computer cluster, where servers or even desktop-class PCs are harvested for distributing the load of multiple long-running programs across large numbers of computers so they can be processed in parallel. A primary outcome of this project is a novel technology that enables simple creation and deployment of "virtual clusters" across the Internet through the use of virtual machine software technologies (such as those provided by commercial and open-source programs including VMware, VirtualBox, Xen, KVM), coupled with zero-configuration virtual networks. In addition to developing and integrating the technology that makes such virtual clusters feasible, this project has deployed Archer, an infrastructure supporting virtual clusters across several universities in the United States - University of Florida, Florida State University, Northwestern University, Northeastern University, University of Texas at Austin, University of Minnesota, and University of Wisconsin-Madison. Through Archer, the project has demonstrated the ability to support research and education in next-generation computer architectures by providing easy access to a distributed computing infrastructure for simulation. In the domain of computer architecture, researchers often need to execute large numbers of long-running simulations in order to quantitatively estimate performance and power consumption improvements that come from enhancements that are proposed for various sub-systems of a computer - such as the micro-processor and memory subsystems. Archer has provided an open computing platform for research in computer architecture leading to results published in conferences and journals, as well as for education through its use in graduate and undergraduate classes. The project web site is at www.archer-project.org, and the infrastructure is available for use by researchers and educators upon request of a user account.
