W. Chou, A.W. Bragg, and A.A. Nilson. The Need for Adaptive Routing in the Chaotic and Unbalanced Traffic Environment. IEEE Transactions on Communications, 29:481--490, April 1981.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... 7 11 16 20 7 5 Hotspot node (College Park) Transient failure link Figure 1: NSFNET T1 Backbone (14 nodes, 21 bidirectional links, average degree 3) Assumptions and parameters used in this paper were predetermined either consistently with those made in the literature, e.g. 15] 12] 9] [5], or from statistics provided by Merit NSFNET Information Services, or by simulations with MaRS. Physical Network Link propagation delays in milliseconds are indicated in Figure 1. Also indicated is the node which will be a hotspot in some of our simulations, and an incident link which will be ....

W. Chou, A.W. Bragg, and A.A. Nilson. The Need for Adaptive Routing in the Chaotic and Unbalanced Traffic Environment. IEEE Transactions on Communications, 29:481--490, April 1981.

....in TCP, the interaction would be more complicated, and we believe it would be hard to draw conclusions about the performance of the routing algorithm itself. Assumptions and parameters used in this paper were predetermined either consistently with those made in the literature, e.g. 15] 12] 9] [5], or from statistics provided by Merit NSFNET 5 9 7 13 9 7 7 7 15 9 8 5 4 14 5 8 7 11 16 20 7 5 Hotspot node (College Park) Transient failure link Figure 1: NSFNET T1 Backbone (14 nodes, 21 bidirectional links, average degree 3) Information Services, or by simulations with ....

W. Chou, A.W. Bragg, and A.A. Nilson. The Need for Adaptive Routing in the Chaotic and Unbalanced Traffic Environment. IEEE Transactions on Communications, 29:481--490, April 1981.

....[20] and its new link cost function (i.e. hop normalized delay function) 18] The tests showed that the new algorithm and cost function give better performance than the old ARPANet al..gorithm (i.e. Distributed Bellman Ford) and the old delay function. Regarding simulation approaches, reference [6] concluded that an adaptive strategy is needed in a skewed workload environment. For SPF, it investigated the effects of the link cost function parameters and demonstrated optimal settings. In [25] the original and new ARPANet al..gorithms (i.e. Distributed Bellman Ford and SPF) were evaluated in a ....

....of the routing algorithm itself. In the rest of this section, we describe the range of parameters exercised in our simulations and the performance measures obtained. Assumptions and parameters were predetermined either consistently with those made in the literature, e.g. 20] 18] 14] [6], or from statistics provided by Merit NSFNET Information Services, or by experimentation. Physical Network Figure 1 illustrates the NSFNET topology. Link propagation delays in milliseconds are indicated. Also indicated is the node which will be a hotspot in some of our experiments. 9 7 13 ....

Wushow Chou, Arnold Bragg, and Arne Nilsson. The need for adaptive routing in the chaotic and unbalanced traffic. IEEE Transactions on Communications, 1981.

....and its new link cost function (i.e. hop normalized delay function) 21] The tests showed that the new algorithm and link cost function give better performance than the old ARPANet al..gorithm (i.e. Distributed Bellman Ford) and the old delay function. Regarding simulation approaches, reference [7] concluded that an adaptive strategy is needed in a skewed workload environment. For SPF, it investigated the effects of the link cost function parameters and demonstrated optimal settings. In [35] the original and new ARPANet al..gorithms (i.e. Distributed Bellman Ford and SPF) were evaluated in a ....

....window. Such packets are temporarily buffered in a produce window . 6 Simulation Parameters In this section, we describe the range of parameters exercised in our simulations. Assumptions and parameters were predetermined either consistently with those made in the literature, e.g. 24] 21] 16] [7], or from statistics provided by Merit NSFNET Information Services, or by simulations with MaRS. We consider the routing algorithms, SPF, MS and ExBF, on the topology of the NSFNETT1 Backbone, for varying workloads, varying link cost functions, varying link failures, and for two types of hardware ....

[Article contains additional citation context not shown here]

W. Chou, A.W. Bragg, and A.A. Nilson. The Need for Adaptive Routing in the Chaotic and Unbalanced Traffic Environment. IEEE Transactions on Communications, 29:481--490, April 1981.

....expected, adaptive schemes perform better when traffic is unpredictable. A two level approach can improve performance by using fixed deterministic routing rules to determine a number of paths between sourcedestination pairs and an adaptive scheme to choose a path based on local traffic variations [92]. This strategy must be made insensitive to small variations in traffic conditions. A similar proposal, also using a two level adaptive routing strategy, utilises a global algorithm to determine a set of good paths between source destination pairs, rather than the best path. These paths are ....

W. Chou, A. Bragg, and A. Nilsson, "The need for adaptive routing in the chaotic and unbalanced traffic environment," IEEE Transactions on Communications, vol. COM-29, no. 4, pp. 481--490, April 1981.

....[19] and its new link cost function (i.e. hopnormalized delay function) 17] The tests showed that the new algorithm and cost function give better performance than the old ARPANet al..gorithm (i.e. Distributed Bellman Ford) and the old delay function. Regarding simulation approaches, reference [4] concluded that an adaptive strategy is needed in a skewed workload environment. For SPF, it investigated the effects of the link cost function parameters and demonstrated optimal settings. In [26] the original and new ARPANet al..gorithms (i.e. Distributed Bellman Ford and SPF) were evaluated in a ....

....Environment In this section, we give a brief overview of MaRS. We then describe the range of parameters exercised in our simulations and the performance measures obtained. Assumptions and parameters were predetermined either consistently with those made in the literature, e.g. 19] 17] 12] [4], or from statistics provided by Merit NSFNET Information Services, or by experimentation. 4.1 Overview of MaRS Our simulation studies were done on a recently developed discrete event simulator, MaRS (Maryland Routing Simulator) It provides a flexible platform for the evaluation and comparison ....

[Article contains additional citation context not shown here]

Wushow Chou, Arnold Bragg, and Arne Nilsson. The Need for Adaptive Routing in the Chaotic and Unbalanced Traffic. IEEE Transactions on Communications, 1981.

....although it avoids deadlock, can increase traffic jams, especially in heavily loaded networks. In order to avoid congested regions of the network, an adaptive routing algorithm can be used. Adaptive strategies have been shown to outperform deterministic strategies in store andforward switching [5], in packet switched communications [24, 35] and in wormhole switching [11, 12] When adaptive routing is used, deadlocks can be avoided in virtual cut through and store and forward switching by misrouting packets in the presence of congestion. This technique, known as deflection or hot potato ....

W. Chou, A.W. Bragg, and A.A. Nilsson, "The need for adaptive routing in the chaotic and unbalanced traffic environment," IEEE Trans. Commun., vol. COM-29, no. 4, pp. 481--490, April 1981.

....to destination, and (2) adaptive routing that allows more freedom in selecting message paths. Most commercial multicomputers use deterministic routing because of its deadlock freedom and ease of implementation. However, adaptive routing can reduce network latency and increase network throughput [4]. It can also tolerate more faults than deterministic routing. But, the flexibility of adaptive routing may cause deadlock and or livelock problems. A deadlock occurs when a message waits for an event that will never happen. In contrast, a livelock keeps a message moving indefinitely without ....

W. Chou, A. W. Bragg, and A. A. Nilsson, "The need for adaptive routing in the chaotic and unbalanced traffic environment," IEEE Trans. on Communications, pp. 481--490, Apr. 1981.

....it avoids deadlock, can increase traffic jams, especially in heavily loaded networks with long messages. In order to avoid congested regions of the network, an adaptive routing algorithm can be used. Adaptive strategies have been shown to outperform static strategies in store and forward routing [3] and in packet switched communications [20, 24] In general, adaptive routing needs additional hardware support. Several adaptive algorithms have been developed for wormhole routing. A deadlock free adaptive algorithm for the hypercube is the Hyperswitch algo4 rithm [4] which is based on ....

W. Chou, A.W. Bragg, and A.A. Nilsson, "The need for adaptive routing in the chaotic and unbalanced traffic environment," IEEE Trans. Commun. , vol. COM-29, no. 4, pp. 481--490, Apr. 1981.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC