A Survey on Core Switch Designs and Algorithms
Dinil Mon Divakaran∗ , Sebastien Soudan† , Eitan Altman‡ , Pascale Vicat-Blanc Primet§ Corresponding author, INRIA / Universit´ de Lyon / ENS Lyon, firstname.lastname@example.org e † ENS Lyon / Universit´ de Lyon / INRIA, email@example.com e ‡ INRIA, firstname.lastname@example.org § INRIA / Universit´ de Lyon / ENS Lyon, email@example.com e
Abstract—Tremendous efforts have gone into research on switch designs and architectures. This survey sketches the evolution of the modern switch architectures, covering the literature over the period 1987-2009. Starting with the simple crossbar switch, we explore various architectures such as output queued, input queued, combined inputoutput queued, buffered crossbar etc., that have evolved during this period. We discuss the pros and cons of these designs, so as to shed light on the path of evolution of switch architecture, and take lessons that will be useful in designing next-generation switches.
I. I NTRODUCTION One of the success factors of the Internet can be attributed to the packet switching technology, as it enables any device to communicate to virtually any number of peers at the same time. At the time data communication networks was ﬁrst designed, the prevailing paradigm for electronic communications was circuit as in telephone network. The new concept used in data networks was introduced by L. Kleinrock in as early as 1960s. Information is ﬁrst divided into small messages — or packets — which are sent over the network and then reassembled at destination. While the end-hosts in a packet-switched network put messages into multiple packets and send them through their interfaces, the network multiplexes packets coming from different end-hosts at the interconnects. These interconnects, or packet switching devices, basically store, route and forward packets before they reach the destination. One core functionality of such switches (a layer...