A high‐performance switch fabric for integrated circuit and packet switching

Abstract

This paper describes an architecture for a high-performance switching fabric that can accommodate circuit-switched and packet-switched traffic in a unified manner. The switch fabric is self-routeing and uses fixed-length minipackets within the switching fabric for all types of connections. Its kernel architecture is based on a routeing topology with individual connection paths from all inputs to all outputs and with FIFO queuing at each output. Owing to the disjoint connection paths, there is no internal blocking, and because of output queueing, output port blocking is prevented to a great extent. The uniformity in architecture allows construction of any size fabric from a single basic module which could be realized on a single chip. Larger-size configurations can be realized either as single-stage or multistage configuration. The second part of this paper discusses performance aspects and gives results and dimensioning guidelines for both circuit-switched and packet-switched traffic.KEY WORDS Integrated circuitlpacket switching Integrated switch fabric Minipackets Self-routeing Output queuing Priorities Single-stage switch Multistage switch Back-pressure 08943222/89/040277-11$05 . SO

Prizma switch technology

“…The attractiveness of IQ lies in its simplicity and low cost: The queues are only required to support a throughput roughly equal to the line speed, whereas in the OQ case each queue must be able to accept the aggregate rate of all inputs. In the early days of fast packet switching, however, performance was the reason why many designs adopted the OQ concept in spite of the more complex and expensive multiport buffers required (e.g., the Bell Laboratories Knockout [5], the NEC Atom [6], and the Siemens Sigma as well as the IBM high-performance switch fabric [7] and Prizma switch [8].…”

“…In this arrangement, the resulting 2N ϫ 2N switch fabric retains its full single-stage characteristic. Because the output queues of two switch chips are actually used in parallel to feed one output port, the resulting performance is even better than with a single chip [25]. A disadvantage is the quadratic growth characteristic: A doubling of the number of ports requires a quadrupling of the number of chips required.…”

“…For larger switches, however, a multi-stage configuration must be chosen. The singlestage expansion and the auto-routing mechanisms used for multi-stage expansion have been inherited from an earlier switch project, known as the "tree switch" [7,26]. In real networks, there exist links of a variety of speeds.…”

A survey of modern high-performance switching techniques

“…Another switch fabric, the design of which is based on an interconnection structure with no internal blocking and queueing capability at the outputs of it modules, is the Integrated Switch Fabric proposed in [53]. This switch fabric is designed to handle both circuit-switched and packet-switched traffic in a unified manner.…”

SXmin: a self-routing high-performance ATM packet switch based on group-knockout principle

“…Several asynchronous transfer mode (ATM) switching networks have been proposed previously 5]. They either require large amounts of hardware and are costly 6,7] or require large internal speedup to reduce packet loss due to blocking 8,9]. Many of these switching networks are based on output queueing because of the higher throughput performances as compared to input queueing networks (approximately 1.72 times that of the input queueing type 10]).…”