Project description

The purpose of the Esprit-funded ARCHES II project was to bridge the gap for the utilization of existing high speed network technologies (HIC) developed in various EC projects for the emerging standard of Gigabit Ethernet (GE). New, scalable switches based on the HIC technologies were designed and put to the test. The role of the University of Amsterdam was to validate the use of GE (for High Performance Computing). As such, the UvA worked as an early adapter of GE technology and a scientific application center. The UvA took up this assignment by looking past the more traditional evaluation and has used both basic benchmark measurements to investigate the various factors influencing the performance and with experiences with future applications (rather than speed-up of today’s applications) which should benefit from this technology. The UvA has developed an application in this context based on the growing importance of scalable web-based information systems. The resulting application ties in with the plans within the Jera and Archipel projects.

The results from the project showed that the technology itself operates as expected, where some weaknesses of the hardware were brought up, such as stability and routing capabilities. However in a wider context, we showed that real-world applications hardly benefit from the technology as it currently stands. The performance is limited by issues that play a role at all levels of the networking infrastructure. This includes PC-hardware, drivers, operating system, communication libraries and programming paradigms. No single component can generally be shown to be the most limiting factor in the poor utilization of very high speed networks in host-to-host communication.