C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A loop-free Bellman-Ford routing protocol without bouncing effect. In ACM SIGCOMM '89, pages 224--237, September 1989.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....the routing information (e.g. routing table) and the functions to update these data structures and propagate their information. In the current version, three types of routing algorithms are implemented: SPF (Shortest Path First) McQuillan et al. 1980] SEGALL [Merlin and Segall, 1979] and ExBF [Cheng et al. 1989]. SPF is a link state routing algorithm, while SEGALL and ExBF are distance vector routing algorithms. Each algorithm uses next hop routing. The user can choose any one of the three routing algorithms; once the choice is made, all the routing components have to be of that type. In order to ....

....only next hop routing is used. 9 In the current version, we have implemented three types of routing components, corresponding to three routing algorithms: SPF, which is very much like the Arpanet routing algorithm [McQuillan et al. 1980] SEGALL [Merlin and Segall, 1979, Segal, 1981] and ExBF [Cheng et al. 1989]. SPF (Shortest Path First) is a link state routing algorithm. In link state algorithms, each node attempts to maintain a database describing the network topology and the link costs. Using this database, each node independently calculates shortest paths, and uses them to determine the next hop ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A loop-free Bellman-Ford routing protocol without bouncing effect. In ACM SIGCOMM '89, pages 224--237, September 1989.

....reflect the link up down status and the traffic flowing through the link. By using these link costs, a routing protocol adapts its next hops to account for topology changes, i.e. failures, repairs and link cost changes. The responsiveness of adaptive routing protocols has received much attention [16, 10, 21, 4, 2, 23]. These studies usually consider a succession 1 of topology changes that leaves the system in a state, say s 0 , where the routes are not shortest. Assuming that there are no further topology changes, they obtain the number of messages (message complexity) and the number of successive hops of ....

....We consider uniform and hotspot with background workloads. Some very preliminary results for Poisson failures and repairs and uniform workload was reported in [20] We consider three routing protocols: Shortest Path First (SPF) 15] Merlin Segall (MS) 16, 19] and Extended Bellman Ford (ExBF) [4]. SPF is a link state routing algorithm, whereas MS and ExBF are distance vector routing algorithms. They are described in section 2. The space requirement at each node is O(N 2 ) for SPF, and O(N Theta e) for MS and ExBF, where N is the number of nodes in the network and e is the average ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

....the source DQDB BMN receives the confirm message, it can migrate the traffic from the old path to the new one. 2. 2 Drawbacks of the DPBB Scheme The distance vector routing method, on which the DPBB scheme is based, is well known to have the drawback of forming short lived and long lived loops [5]. This is because nodes choose their next hops in a completely distributed fashion based on possibly stale or even incorrect information. There are loop avoidance techniques available in the literature; they normally require inter node coordination. The DPBB scheme alleviates the loop as well as ....

.... more efficient, easier to implement and requires less storage space than the link state approach, it is well known that it can cause the formation of routing loops because each node updates its routing table in a distributed fashion based on possibly stale or even incorrect information [5]. When applied to a cluster of DQDBs, the distance vector ap9 DQDB STATUS DQDB node bridge node DQDB Status Table Cluster Graph Figure 2: Each DQDB BMN maintains the DQDB status table and the cluster graph proach naturally resulted in a scheme like DPBB [13] As discussed in the previous ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-Free BellmanFord Routing Protocol Without Bouncing Effect. In ACM SIGCOMM, pages 224--237, Sept. 1989.

....reflect the link up down status and the traffic flowing through the link. By using these link costs, a routing protocol adapts its next hops to account for topology changes, i.e. failures, repairs and link cost changes. The responsiveness of adaptive routing protocols has received much attention [16, 10, 20, 4, 2, 22]. These studies usually consider a succession 1 of topology changes that leaves the system in a state where the routes are not shortest. Assuming that there are no further topology changes, they obtain the number of messages (message complexity) and the number of successive hops of message ....

....it has no workload and assumes unit link delays and zero processing time at the nodes. 1 Poisson failures and repairs and uniform workload was reported in [19] We consider three routing protocols: Shortest Path First (SPF) 15] Merlin Segall (MS) 16, 18] and Extended Bellman Ford (ExBF) [4]. SPF is a link state routing algorithm, whereas MS and ExBF are distance vector routing algorithms. They are described in section 2. The space requirement at each node is O(N 2 ) for SPF, and O(N Theta e) for MS and ExBF, where N is the number of nodes in the network and e is the average ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

....link delay. Experience showed that this leads to unstable oscillations, and in 1987 it was replaced by a more slowly changing hop normalized delay function, which uses exponential averaging, movement limit, etc [18] Since 1979, many new kinds of distance vector algorithms have been proposed [21, 15, 5, 13, 23] which achieve significantly reduced dissemination times by using node coordination mechanisms. The Merlin Segall algorithm [21, 24] has a worst case dissemination time of O(D 2 ) hop count. The Jaffe Mosse algorithm [15] the DUAL algorithm by Garcia Luna Aceves [13] and the Extended ....

....dissemination times by using node coordination mechanisms. The Merlin Segall algorithm [21, 24] has a worst case dissemination time of O(D 2 ) hop count. The Jaffe Mosse algorithm [15] the DUAL algorithm by Garcia Luna Aceves [13] and the Extended Bellman Ford algorithm by Cheng et al. [5] all have worst case dissemination time of O(D) hop count. Although the last three algorithms have the same order of dissemination times in terms of hop count, in terms of real time they have successively smaller dissemination times because they impose successively weaker coordination ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. GarciaLuna -Aceves. A loop-free Bellman-Ford routing protocol without bouncing effect. In ACM SIGCOMM '89, pages 224--237, September 1989.

....functions. For modeling the physical nodes and links of the target system, MaRS has link components and node components. For modeling routing protocols, MaRS has a link cost function component and three routing components, SPF, ExBF, and SEGAL, corresponding to three different routing algorithms [12, 6, 13, 14]. For modeling application and transport protocols (e.g. user traffic and flow control) MaRS has three types of workload components, FTP, TELNET, and Simple Traffic, corresponding to three different types of user traffic. Each type of workload component has two versions, a source and a sink. MaRS ....

....it is added to the appropriate poll lists depending on its type. 7 Concluding Remarks We have found MaRS to be very useful in understanding the complex dynamics of routing algorithms. In [16] we compared the performance of two new distance vector routing algorithms, SEGAL [13, 14] and ExBF [6], and a link state algorithm, SPF [12] We considered the NSFNETT1 backbone network with varying uniform and skewed workloads and failure distributions. Our overall conclusion was that ExBF is as good as SPF for good link cost functions, whereas SEGAL is worse. In [3] we examined the ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

....was link delay. Experience showed that this leads to unstable oscillations, and in 1987 it was replaced by a more slowly changing hopnormalized delay function, which uses exponential averaging, movement limit, etc [21] Since 1979, many new kinds of distance vector algorithms have been proposed [25, 30, 19, 6, 15, 29] which achieve significantly reduced dissemination times by using node coordination mechanisms. The Merlin Segall algorithm [25, 30] has a worst case dissemination time of O(H 2 ) hop count, where H is the length of a maximum length shortest path between any two nodes. The worstcase ....

....of O(H 2 ) hop count, where H is the length of a maximum length shortest path between any two nodes. The worstcase dissemination time is O(H) for the Jaffe Mosse algorithm [19] and the DUAL algorithm by Garcia Luna Aceves [15] and is O(N) for the Extended Bellman Ford algorithm by Cheng et al. [6]. However, in terms of real time they may have successively smaller dissemination times because they impose successively weaker coordination constraints. The obvious question is: Are the new distance vector algorithms good enough to replace SPF To answer this, we must compare these algorithms in ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

....functions. For modeling the physical nodes and links of the target system, MaRS has link components and node components. For modeling routing protocols, MaRS has a link cost function component and three routing components, SPF, ExBF, and SEGAL, corresponding to three different routing algorithms [8, 9, 10, 11]. For modeling application and transport protocols (e.g. user traffic and flow control) MaRS has three types of workload components, FTP, TELNET, and Simple Traffic, corresponding to three different types of user traffic. Each type of workload component has two versions, a source and a sink. MaRS ....

....the complex dynamics of routing algorithms. As described earlier, a routing algorithm may form loops in its routes while adapting to topology changes, which degrades performance. In [1, 17] we compared the performance of two new distance vector routing algorithms, SEGAL [10, 11] and ExBF [9], and a link state algorithm, SPF [8] SPF and ExBF allow short lived loops, whereas MS avoids loops entirely. In response to a workload or topology change, to avoid loops MS starts a diffusion computation during which nodes keep their old routes until they obtain new loop free routes. In case ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

.... loops giving rise to large dissemination times [2] The ARPANET initially used this Distributed Bellman Ford algorithm, but it was replaced in 1979 by a link state algorithm referred to as SPF (Shortest Path First) 19] Since 1979, many new kinds of distance vector algorithms have been proposed [20, 24, 15, 3, 11, 23] which achieve significantly reduced dissemination times by using node coordination mechanisms. We do not know whether the new distance vector algorithms are good enough This work is supported in part by RADC and DARPA under contract F30602 90 C 0010 to UMIACS at the University of Maryland, ....

....throughput under varying file transfer workload. In all cases, we consider the NSFNET backbone network and a hop normalized link cost function. We compare SPF and two recently proposed distance vector algorithms, namely Merlin Segall [20, 24] henceforth called MS, and Extended Bellman Ford [3] henceforth called ExBF. The file transfer workload we consider is specified in terms of source sink pairs. We consider two kinds of geographic distributions for the source sink pairs: uniform, where the pairs are distributed uniformly among the nodes of the network, and ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

.... but because of long lived loops, it was replaced in 1979 by a brute force link state algorithm which requires O(N 2 ) space at each node (to maintain a view of the network topology with a cost for each link) Since 1979, many new kinds of distance vector algorithms have been proposed [9, 6, 12, 2, 5, 11] which avoid long lived loops by using various node coordination mechanisms. For example [9, 6] use diffusion computations [3] to avoid loops entirely. References [12, 2, 11] avoid long lived loops, but not short lived loops, i.e. loops that disappear in time proportional to N or less. In [12] ....

....topology with a cost for each link) Since 1979, many new kinds of distance vector algorithms have been proposed [9, 6, 12, 2, 5, 11] which avoid long lived loops by using various node coordination mechanisms. For example [9, 6] use diffusion computations [3] to avoid loops entirely. References [12, 2, 11] avoid long lived loops, but not short lived loops, i.e. loops that disappear in time proportional to N or less. In [12] each node maintains for each destination a set of paths (in addition to the distances) one for each of its neighbors. The intention is that the path maintained at node u for ....

[Article contains additional citation context not shown here]

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free BellmanFord Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

....the classical Distributed Bellman Ford (DBF) algorithm [2] Compared to link state method, it is computationally more efficient, easier to implement and requires much less storage space. However, it is well known that this algorithm can cause the formation of both short lived and long lived loops [3]. The primary cause for formation of routing loops is that nodes choose their next hops in a completely distributed fashion based on information which can possibly be stale and, therefore, incorrect. Almost all proposed modifications to DBF algorithm [6, 7, 9] eliminate the looping problem by ....

.... operational routing protocols which employ these complex coordination methods to achieve loop freedom, which leads us to the conclusion that from a practical Routing Method Looping Internodal Space Coordination Complexity Bellman Ford [2] s l O(nd) Link State [8] s O(n 2 ) Loop free BF [3] s O(nd) RIP [5] s l O(n) Merlin Segall [9] loop free Required O(nd) Jaffe Moss [6] loop free Required O(nd) DSDV loop free O(n) s short term loop, l long term loop n number of nodes, d maximum degree of a node Table 5: Comparison of various routing methods point of view the ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A loop-free Bellman-Ford routing protocol without bouncing effect. In ACM SIGCOMM '89, pages 224--237, September 1989.

.... but because of long lived loops, it was replaced in 1979 by a brute force link state algorithm which requires O(N 2 ) space at each node (to maintain a view of the network topology with a cost for each link) Since 1979, many new kinds of distance vector algorithms have been proposed [15, 19, 10, 21, 3, 6, 17, 9] which avoid long lived loops by using various node coordination mechanisms. For example [15, 19, 10] use diffusion computations [4] to avoid loops entirely. References [21, 3, 17, 9] avoid long lived loops, but not short lived loops, i.e. loops that disappear in time proportional to N or ....

....a cost for each link) Since 1979, many new kinds of distance vector algorithms have been proposed [15, 19, 10, 21, 3, 6, 17, 9] which avoid long lived loops by using various node coordination mechanisms. For example [15, 19, 10] use diffusion computations [4] to avoid loops entirely. References [21, 3, 17, 9] avoid long lived loops, but not short lived loops, i.e. loops that disappear in time proportional to N or less. In [21] each node maintains for each destination a set of paths (in addition to the distances) one for each of its neighbors. The intention is that the path maintained at node u ....

[Article contains additional citation context not shown here]

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. Garcia-Luna-Aceves. A Loop-free Bellman-Ford Routing Protocol Without Bouncing Effect. In Proc. ACM SIGCOMM '89, pages 224--237, September 1989.

....A link models a bidirectional transmission channel between two nodes, whose parameters include bandwidth, propagation delay and failure and repair distributions. MaRS supports several routing algorithms, including SPF (Shortest Path First) 19] Merlin Segall [20] and Extended Bellman Ford [9]. An associated link cost component maintains and updates the cost of each link. The cost functions can be based upon hop count, utilization or delay. The workload in MaRS is defined in terms of source sink component pairs that are attached to nodes. There are three types of workload in MaRS, ....

Cheng, C., Riley, R., Kumar, S.P.R., and Garcia-Luna-Aceves, J.J. A loop-free Bellman-Ford routing protocol without bouncing effect. In Proceedings of ACM SIGCOMM'89, pages 224--237, September 1989.

....the classical Distributed Bellman Ford (DBF) algorithm [2] Compared to link state method, it is computationally more efficient, easier to implement and requires much less storage space. However, it is well known that this algorithm can cause the formation of both short lived and long lived loops [3]. The primary cause for formation of routing loops is that nodes choose their next hops in a completely distributed fashion based on information which can possibly be stale and, therefore, incorrect. Almost all proposed modifications to DBF algorithm [6, 7, 9] eliminate the looping problem by ....

....proved by Jaffe and Moss [6] which states that in presence of static or decreasing link weights distance vector algorithms always maintain loop free paths. Routing Method Looping Internodal Space Coordination Complexity Bellman Ford [2] s l O(nd) Link State [8] s O(n e) Loop free BF [3] s O(nd) RIP [5] s l O(n) Merlin Segall [9] loop free Required O(nd) Jaffe Moss [6] loop free Required O(nd) DSDV loop free O(n) s short term loop, l long term loop n number of nodes, d maximum degree of a node Table 5: Comparison of various routing methods 6 Comparison with ....

C. Cheng, R. Riley, S. P. R. Kumar, and J. J. GarciaLuna -Aceves. A loop-free Bellman-Ford routing protocol without bouncing effect. In ACM SIGCOMM '89, pages 224--237, September 1989.

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