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

Practical algorithms for multicast support in input queued switches

“…We use two previously proposed multicast scheduling algorithms, named RS (Random Scheduler) and GMSS (Greedy Min Split Scheduler) [5], the former being chosen for its simplicity as a reference case, the latter due to its good performance. To satisfy the in-order cell delivery constraint, when a multicast cell is partially served, it remains at the head of its queue, and will contend for the residual set of destinations in the next time slot.…”

“…We denote by N c the cardinality of the set of all generated fanout sets, and by P f the probability of generating a fanout set with fanout f , 1 ≤ f ≤ N . The considered distribution for the fanout sets is named binomial fanout, which was proven to be very difficult to schedule [5] and highlights disciplines differences: the fanout set is chosen according to a non-uniform binomial distribution, with mean fanout h m . Hence, N c = 2 N − 1 and…”

“…It was already shown in [5] that it is very important to focus on gathered traffic, i.e., traffic gathered over few active input ports (M ≤ N ) and equally distributed over all output ports. Indeed, when M = N and inputs are equally loaded, the maximum sustainable traffic leads to a load at each input which is at most 1/h m .…”

“…We assume that data in-order delivery is a requirement that is directly supported in the switch, avoiding any reordering at output ports; as a consequence, we cannot rely on cell-bycell load balancing schemes (if not as a reference case), but we need to define queueing disciplines, i.e., assignments of multicast flows to queues. Previous work includes [4], [5], where performance of IQ switch with k queues per input port, with k = O(N ), are discussed. Both queueing and scheduling disciplines are defined and compared by simulation.…”

Multicast Due Date Round-Robin Scheduling Algorithm for Input-Queued Switches

“…One such practical queuing scheme utilized for multicast switches is to assign a single FIFO queue at each input for all multicast traffic, however, the HOL blocking quandary limits the throughput. Whereas the other algorithms [1], [3] considered a circumscribed number of FIFO queues is maintained at each input to reduce the HOL blocking problem. Thus queuing architecture is denominated as k-MC-VOQ and performance of these multicast switches are analyzed theoretically [12], [15].…”