D.M. Chiu R Jain/ Congestion Avoidance in Camputer Netwarks hardware and software scales or parameters as much as possible. The reason being that in a guire human help to configure. having algorithms depend on system scales would complicate the configuration task and make the algorithm more Alloc vuInerable to human crror. while in this report we have discussed optimizing the control scheme ccording to certain criteria. practical considera- on may dictate that the control be chosen for the widest range of values of system parameters. As we indicated earlier. the nonlinear controls tend to be more sensitive to the system parameters and hus, are less likely to be useful in practice Uger 1'g Alloeation xt Another possible constraint is that the resource and thus the allocations must be integral. Fe Fig. 8. Vectorial representation for nonlinear controls. example buffers and windows are all measured in integers. Simple rounding off to the nearest integer may cause violation of the various convergence A nonlinear control could generally include conditions more components with different slopes Ease of implementation could also affect the n()=∑a(x(1) choice of the controls. For example, the number of multiplications or exponentiations required to implement a control would impact on the minimal Then the sum of the components must have a hardware require slope satisfying the above condition. Although nonlinear controls offer us far more 5.1. Further Question flexibility in trying to direct towards fairness, it also complicates the task of finding the right There are many further questions worth explor caling factors, represented hy ak in the ahove ng in conjunction to this nrohlem In particula equation. These parameters usually must be cho- the follo ng are impo sen relative to system parameters, such as the (1) How does delayed feedback affect the control? capacity Xgoul and maximum number of users In practice there is invariably some delays before Nmax, Being too sensitive to system parameters the feedback arrives at the controller As the delay educes the robustness of the control. For this lengthens, the feedback becomes less and less use- cal ful, and the perfomance worsens. It would be controls. We will discuss the robustness question valuable to have quantitative assessment of how more in the next section the performance degrades (2)What is the marginal utility of increased bits of feedback? The binary feedback is the simples 5. Practical Considerations Adding additional feedback signals may help to cut down the oscillations. A formal analysis would The problem studied in this report is a generi allow an assessment of the tradeoff of perfor but also highly abstracted problem. In order to mance versus complexity apply the iesults tu solve the deenualized cung ()Is it worthwhile to guess the current number tion control Problem in real networks, many prac- f users n? Uscrs come and go dynamically and tical issues must be taken into considerations We the number changes by integral values. A se- briefly discuss some of them here. quence of increase signals may indicate the reduc One general principle in choosing an algorithm tion in the number of users from n to n-1. If n in a general architecture is to be independent of were known or bounded, sources could predict