Load balanced Birkhoff-von Neumann switches, part II: multi-stage buffering

The paper studies input-queued packet switches loaded with both unicast and multicast traffic. The packet switch architecture is assumed to comprise a switching fabric with multicast (and broadcast) capabilities, operating in a synchronous slotted fashion. Fixed-size data units, called cells, are transferred from each switch input to any set of outputs in one time slot, according to the decisions of the switch scheduler, that identifies at each time slot a set of nonconflicting cells, i.e., cells neither coming from the same input, nor directed to the same output. First, multicast traffic admissibility conditions are discussed, and a simple counterexample is presented, showing intrinsic performance losses of input-queued with respect to output-queued switch architectures. Second, the optimal scheduling discipline to transfer multicast packets from inputs to outputs is defined. This discipline is rather complex, requires a queuing architecture that probably is not implementable, and does not guarantee in-sequence delivery of data. However, from the definition of the optimal multicast scheduling discipline, the formal characterization of the sustainable multicast traffic region naturally follows. Then, several theorems showing intrinsic performance losses of input-queued with respect to output-queued switch architectures are proved. In particular, we prove that, when using per multicast flow FIFO queueing architectures, the internal speedup that guarantees 100% throughput under admissible traffic grows with the number of switch ports.

Multicast traffic in input-queued switches: optimal scheduling and maximum throughput

The paper studies the design of IQ cell-based switch architectures with a reduced number of queues at input ports to support multicast traffic delivery. The design involves the definition of a queueing policy (to select how and where to enqueue packets) and of a scheduling policy (to select when and from which queue to transfer packets, satisfying the transmission constraints given by the switching fabric). The possible tradeoffs between performance figures, queueing policies and scheduling algorithms, are discussed. Clever design can lead to non-negligible performance gains; thus, several queuing policies and scheduling algorithms are examined by simulation. The main contributions of the paper are the following. First, it is shown that a small number of queues is sufficient to obtain good switch performance under typical traffic patterns. Second, multicast traffic patterns difficult to schedule are obtained by concentrating the load over few input ports, while keeping the output port load high. Third, the scheduling algorithm is shown to have a deeper impact on the switch performance than the queuing policy. Fourth, queueing policies that balance the load among all the available queues are shown to perform better than queuing policies based on other criteria. Finally, a novel greedy scheduling policy that improves switch performance is presented.

On the number of input queues to efficiently support multicast traffic in input queued switches

We propose an efficient multicast cell-scheduling algorithm, called multiple-slot cell-scheduling algorithm, for multicast ATM switching systems with input queues. Cells in an input-queueing system are usually served based on the first-in-first-out (FIFO) discipline, which may have a serious head-of-line (HOL) blocking problem. Our algorithm differs from previous algorithms in that we consider the output contention resolution for multiple time slots instead of the current time slot only. Like a window-based scheduling algorithm, our algorithm allows cells behind an HOL cell to be transmitted prior to the HOL cell in the same input port. Thus, HOL blocking can be alleviated. We have illustrated that the delay-throughput performance of our algorithm outperforms most of those algorithms that consider only the output contention resolution for the current time slot. We also present a simple and efficient architecture for realizing our algorithm, which can dramatically reduce the time complexity. We believe that the proposed architecture is very suitable for multicast asynchronous transfer mode (ATM) switching systems with input queues.

An efficient cell-scheduling algorithm for multicast ATM switching systems

In this paper we discuss the throughput achievable in input-queued cell-based switches loaded with multicast traffic. The switch architecture is assumed to comprise a synchronous broadcast switching fabric, where fixed-size data units, called cells, can be transferred in one slot from one Input to any set of outputs. The switch scheduler must select the time slots for transfers of non-conflicting cells, i.e., cells neither coming from the same input nor directed to the same output. Contrary to the case of unicast traffic, for which input-queued switches were proved to yield the same throughput as output queued switches, we show by simulation experiments and analytical modeling that throughput limitations exist in input-queued switches loaded with multicast traffic.

On the throughput of input-queued cell-based switches with multicast traffic

The authors consider multicast packet switching where an input may send the same packet to many outputs within an ATM (asynchronous transfer mode) time slot. A host of multicast queuing disciplines can be exercised. Assuming only independent HOL (head of line) service to an output from slot to slot, the delay performance and saturation throughput are derived. The accuracy of the assumption for different disciplines is examined via extensive simulation. It is shown that the FCFS (first come, first serve) HOL service discipline not only has almost completely identical results for simulation versus analysis, but also provides the best saturation throughput, fairness, and delay performance among all disciplines considered. The analysis shows that implementing packet priorities can significantly improve delay performance. It is concluded that offering priority according to age in the queue is a worthwhile feature for multicast (as well as unicast) packet switches. >

T. Renner

Papers

Queueing strategies for multicast packet switching