Scheduling in Non-Blocking Buffered Three-Stage Switching Fabrics

Abstract

Abstract-Three-stage non-blocking switching fabrics are the next step in scaling current crossbar switches to many hundreds or few thousands of ports. Congestion management, however, is the central open problem; without it, performance suffers heavily under real-world traffic patterns. Schedulers for bufferless crossbars perform congestion management but are not scalable to high valencies and to multi-stage fabrics. Distributed scheduling, as used in buffered crossbars, is scalable but has never been scaled beyond crossbar valencies. We combine ideas from central and distributed schedulers, from request-grant protocols and from credit-based flow control, to propose a novel, practical architecture for scheduling in non-blocking buffered switching fabrics. The new architecture relies on multiple, independent, single-resource schedulers, operating in a pipeline. It: (i) isolates well-behaved against congested flows; (ii) provides throughput in excess of 95% under unbalanced traffic, and delays that successfully compete again output queueing; (iii) provides weighted max-min fairness; (iv) directly operates on variable-size packets or multi-packet segments; (v) resequences cells or segments using very small buffers; and (vi) can be realistically implemented for a 1024×1024 reference fabric made out of 32×32 buffered crossbar switch elements. This paper carefully studies the many intricacies of the problem and the solution, discusses implementation, and provides performance simulation results.

Analysis of the Effects of XLFrames in a Network

“…The reduction in the number of packets also reduces the number of arbitration decisions required to switch packets from the inputs to the outputs. Though this is not possible in traditional switches using cell-based switching (where variable length packets are segmented into fixed size cells), it should be noted that with the feasibility of adding buffers in a crossbar switch, recent research works have explored asynchronous buffered crossbar architecture, proposing it as the next level of crossbar switches that can scale to hundreds of ports [17,18]. As the units being switched in such architectures are variable-size packets, the gain due to XLFs is directly proportional to the reduction in the number of packets.…”

Scheduling in switches with small internal buffers

“…Our model is crude because it assumes that all buffers are at the outputs; however, it constitutes a useful first approximation towards a full study of the fabric itself. We are undertaking such a full study in a subsequent, current work [11]. To achieve the above small buffer spaces, we had to go over a sequence of steps which we present in this paper.…”

Congestion management for non-blocking clos networks

“…In [9], we proposed an alternative congestion management for non-blocking, 3-stage Clos fabrics. Before injecting a packet into the fabric, input ports have to first request and be granted permission from a control unit.…”

“…This avoids HOL blocking in 2nd stage. Combined with inverse multiplexing, this rule additionally guarantees that intermediate buffers only rarely fill up (only due to unlucky routing), thus preventing HOL blocking throughout the fabric -see [9] or [10,Sec. 4.2].…”