IBM PowerPRS switches
Advances in packet-switching technology have enabled the rapid growth of the Internet and will be crucial for its continued expansion in the future. For almost ten years, the Prizma switch architecture has been a main component of IBM network technology, and today it is a cornerstone of the IBM Microelectronics Division's engagement with leading network product providers. In this paper we review the evolution of the Prizma architecture over the last decade and discuss the PowerPRS range of switch-chip products that are based on the Prizma architecture.
The Prizma packet switch architecture, developed at IBM's Zurich Research Laboratory in the past decade, allows fabrics with a capacity ranging from a few Gigabits to multi-Terabits to be built from small numbers of high-speed switching chips. The Prizma architecture was originally developed specifically for broadband telecommunications and can be used for any high-speed switching or routing function.
The key features of the Prizma switch architecture are very high performance achieved by a strict separation of control and data paths, efficient queuing algorithms, and very large-scale integration on a single chip. This provides a unique level of flexibility, enabling multiple chips to be combined in many ways to build a broad range of communication nodes.
The IBM PowerPRS packet routing switches based on the first and second generation of the Prizma architecture have now been available for several years. The second generation allows a 0.5 Tb/s switch to be built on a single card. This generation combines virtual-output queuing and output buffering and is described in an award-winning paper. At the end of September 2003, IBM sold the PowerPRS business.
The third-generation Prizma architecture represents a major evolution from the second generation in addressing computer interconnect fabric needs.