This paper considers how to construct non-power-of-two distributors, which concentrate and rotate, simultaneously, active input packets for an efficient buffer utilization in packet switching environment. As a theoretical background, we first establish that an R R distributor cannot be constructed out of a conventional N N distributor for a non-power-of-two number R in N/2 < R < N(=2n), due to internal packet conflicts. We then consider how to resolve the internal packet conflict problem to construct a non-power-of-two distributor out of the nearest power-of-two distributor and introduce a number of possible arrangements. Especially, we present a single-CSE (Controlled Switching Element)-RBN (Reverse Banyan Network) based realization which helps to eliminate the dummy address calculation and generation operations that used to be required in the conventional RBN based distributors. Further, we extend the single-CSE-RBN based distributor to generate two flexible-size distributor structures--flexible R R and N R distributors--which can respectively perform the R R and N R distribution functions for an arbitrary number R N simply by changing the CSE parameter setting.

In this paper, we develop parallel scrambling techniques for the distributed sample scrambling (DSS), which are directly applicable to the bit- and multibit-interleaved multiplexing environments. We first consider how to realize PSRGs, parallel samplings and parallel corrections for the multibit-parallel DSS (MPDSS), which are the fundamental problems in realizing the MPDSS scramblers and descramblers. The results are summarized in three sets of theorems, and a corollary is attached to each theorem to specifically handle the case of the parallel DSS (PDSS). The theorems and corollaries are supported by examples that demonstrate the relevant MPDSS scramblers and descramblers.

In this paper, we present the FB (Fair Buffering) mechanism for the efficient support of TCP traffic over UBR connections in the subnet ATM model. We show that both throughput and fairness may be improved by using this congestion control mechanism. The FB mechanism is founded on the observation that the performance of TCP over UBR connections is optimal when the buffer space is allocated in proportion to the connection's bandwidth-delay product. We compare the performance of the existing drop-tail, EB (Equal Buffering) and the proposed FB buffer management schemes, with and without RR (Round-Robin) scheduling, and show through simulation the effectiveness of the proposed FB mechanism when used with the RR scheduling scheme.

For backward compatibility, ECN-capable networks should be capable of handling both ECN-capable and ECN-incapable TCP flows. In this letter, we present a backward congestion notification (BCN) scheme that can provide fast congestion indication delivery, while improving fairness between ECN-capable and ECN-incapable flows. Simulation results reveal that the BCN scheme is more effective than the original ECN mechanism in terms of stability, throughput, and fairness.

This paper analyzes the performance of an ATM cell multiplexer with a two level MMPP input on a discrete-time basis. We approximated the input process as a simple MMPP model. We developed an MMPP/D/1/K queueing model for the ATM cell multiplexer, and employed an analytic approach for the evaluation of cell loss probability. We verified the accuracy of the results using computer simulation. We applied the above analytic method to connection admission control (CAC) of the ATM network. The resulting connection admission control scheme employs the concept of the "effective bandwidth" and table-look-up procedure. We confirmed through a computer simulation that the proposed connection admission control scheme outperforms the peak bandwidth allocation scheme with respect to link utilization.