Abmuct-This is an expanded version of an earlier report, adding, new results. The first part introduces a simple analytic model showing that dynamic routing may increase network throughput or lower delay when trunks are moderately loaded, but will reduce throughput or increase delay at heavy trunk loads. The second part of the paper presents SimukttiOn results characterizing: 1) the operation of a network "Protected" by end-to-end flow control, 2) the operation of an ''unprotWted" network, 3) the transition between these two modes, 4) the effect of changing the update interval for the routing algorithms, 5) the effect Of Slow local loops, and 6 ) the effect of increasing the number of alternate paths available for dynamic routing.Overall, the results show that dynamic routing improves network performance only over a small parameter range, and should dynamic routing be used at all, it must be implemented with great care.
INTRODUCTIONT HE overall congestion-control strategy of a packet-switched network specifies how that network protects itself from potential overload and how network resources are allocated to network users. In [ 11, it was suggested that congestion control has three component subareas: flow control, which prevents an unfair allocation of resources among competing users; routing, which strives to balance the load over equivalent network resources; and scheduling, which can reallocate network delay among various user classes.While the importance of congestion control is clear-a bad strategy will be very apparent to the user-the current understanding of congestion control is very limited. It is hoped that this paper which is an extension of an earlier paper [2] adds to this understanding.First, a simple analytic model is used to show where dynamic routing is useful. Second, the results of a simulation study show that dynamic routing may have adverse effects (worsen performance) in a network protected by flow control and that, at best, the performance of dynamic routing algorithms is quite sensitive to various parameters.Taken together, the results indicate that: 1) it is quite * dangerous to try to study flow control, routing, or scheduling as isolated mechanisms; and 2 ) before any conclusions can be drawn about the performance of a particular set of congestion control algorithms, the overall objectives of a particular network must be very carefully thought out. We shall see that minimization of delay or maximization of throughput require different congestion strategies.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.