Abstract: Flow control is proposed as a means of obtaining an "optimal tradeoff" between low delay and high throughput in computer networks Several versions of "optimal tradeoff" are defined based on network power A class of algorithms which attempt to optimize network performance are investigated These algorithms operate on the design principles of dynamic, distributed execution and use of local information These design principles force the algorithms to be suboptimal, and we thus investigate the relative performance of each in different network configurations Several properties of power as a network performance objective function are examined In certain configurations, two variations of network power are unfair to certain users by not permitting them to send any messages A version of network power ("product of powers") corrects this deficiency Other properties discussed include the nonconvexity of the generalized power function

Abstract: This paper focuses on network delays as they apply to voice traffic. First the nature of the delay problem is discussed and this is followed by a review of enhanced circuit, packet, and hybrid switching techniques: these include fast circuit switching (FCS), virtual circuit switching (VCS), buffered speech interpolation (SI), packetized virtual circuit (PVC), cut-through switching (CTS), composite packets, and various frame management strategies for hybrid switching. In particular, the concept of introducing delay to resolve contention in SI is emphasized, and when applied to both voice talkspurts and data messages, forms a basis for a relatively new approach to network design called transparent message switching (TMS). This approach and its potential performance advantages are reviewed in terms of packet structure, multiplexing scheme, network topology, and network protocols. The paper then deals more specifically with the impact of variable delays on voice traffic. In this regard the importance of generating and preserving appropriate length speech talkspurts in order to mitigate the effects of variable network delay is emphasized. The results indicate that a desirable length of talkspurt "hangover" of about 200 ms will accomplish this without unduly affecting speech activity, and that, under these circumstances, the perceptable threshold of variable talkspurt delay can be as high as about 200 ms average. As such, the results provide a useful guideline for integrated services system designers. Finally, suggestions are made for further studies on performance analysis and subjective evaluation of advanced integrated services systems.

Abstract: Some analytical results have been developed for the analysis and design of a packet-voice receiver in previous work [1], namely, the delay suffered by packets because of the reassembly mechanism is described in terms of the overall delay probability density function (pdf) w(t) . This basic analytical result is used here in order to obtain the pdf b(t) of the time spent by packets in the receiver buffer. In this way, the previous work [1] is extended to analysis of the delay distribution of the receiver buffer where arriving talkspurts are delayed to compensate for the variable network delay of all packets within a talkspurt. The tools developed here facilitate a choice of the buffer size at the prescribed level of packet-loss probability.

Abstract: @, where N is the number of nodes in the network. We show that this network is optimal in terms of hop distance between nodes, delay, throughput, and terminal reliability. The paper includes derivation of closed form expressions for the maximum and average hop distance between nodes, number of distinct routes between two farthest nodes, and throughput. The effect of node and link failures on network performance is also considered.

Abstract: A shared network is largely oversubscribed by independent users who make random demands on the network. Network flow control is required for the orderly operation of the network under all potential traffic loads. This paper presents a qualitative analysis of the theory of flow control to circuit-switched and packet-switched networks and pro- ,. poses 4 complementary controls and 1 subtended control to constitute "flow control." They are traffic control, routing and delay control, congestion control, network management control, and end-end flow control. This paper introduces the concept of the "R-T function" to illustrate the synthesis of delays and routing freedoms, and to demonstrate the continuum between overload and congestion phenomena. The application of the qualitative theory of flow control to the Canadian Telephone Network is reviewed and the agreement between the theory and the practice is demonstrated. The telephone network is a prime example of a circuit-switched network which has received the benefit of extensive simulation and analytical studies, as well as a long experience to validate the conclusions reached in this paper.

Abstract: In a data network, when messages arrive at a switch to be served (transmitted) on a line, it seems reasonable to assume that the arrival process can be described as a Poisson (random) process However, when messages are divided into a number of packets of a maximum length, these packets arrive bunched together This gives rise to what is referred to as “peaked” traffic The degree of peakedness depends on 1) the interarrival time of packets associated with a particular message and 2) the distribution of the number of packets per message In this paper we describe a queueing model which accounts for the non-Poissonian nature of the packet arrival process as a function of these two factors Since packets are of a fixed maximum length, the model assumes that the packet service time is constant, as opposed to the mathematically more tractable but less realistic assumption of exponentially-distributed service time This queueing model is then used to describe the network delay as affected by: 1 Message switching versus packet switching, 2 A priority discipline in the queues, 3 Packet interarrival time per message, which is probably controlled by the line speed at the packet origination point, and 4 A network which carries only short inquiry-response traffic as opposed to a network which also carries longer low-priority printer traffic The general conclusions are that the peakedness in the arrival process caused by a short interarrival time of packets per message and the longer printer traffic would cause excessive delays in a network If inquiry-response traffic with a short response-time requirement is also to be carried on the same network a priority discipline has considerable value Message switching for such a combination of traffic should be avoided

Abstract: In a packet network using medium speed terrestrial connectivity, average end-to-end delay can be reduced by using highspeed satellite overflow channels despite their propagation delay. To investigate networks of this type analytically, we derive a new basic queueing model for overflow systems with buffered traffic. The simple, closed-form expressions facilitate analysis. Approximations are developed for multiple primary systems with a single overflow channel; these agree with simulation. Using the overflow model analysis indicated the following. The reduced queueing delay to the high capacity overflow channel more than compensated for propagation delay. Inherent, broadcast transmission capabilities of satellite channels reduced overall network delay. The high capacity overflow channels permit networks to withstand substantial nodal imbalance and overloads.

Abstract: A linear transmission path of a communication network is considered, where some interfering traffic of random character is assumed. Limit theorems and laws of iterated logarithm are derived for the asymptotic network delay by the help of weak and strong invariance principles.