Elsevier

Computer Communications

Volume 24, Issues 15–16, 1 October 2001, Pages 1673-1684
Computer Communications

Design and analysis of a merging algorithm for multipoint-to-point ABR service in ATM networks

https://doi.org/10.1016/S0140-3664(01)00300-0Get rights and content

Abstract

In this paper, we propose a merging algorithm, which can provide efficient support for multipoint-to-point ABR service in ATM networks. By forwarding the FRM cells belonging to the VC with the largest FRM-cell arrival rate in a merge point, the proposed algorithm can achieve better link utilization than existing merging algorithms. In addition, the proposed algorithm reduces the number of FRM cells forwarded by a merge point. As a result, it can reduce the control overhead of ABR service. Most importantly, it does not incur extra complexity in switches. We also discuss the impact of different network topologies on our algorithm. Simulation results show that the proposed algorithm is able to achieve better performance while requiring significantly fewer RM cells.

Introduction

Multipoint communication is the exchange of information among multiple senders and multiple receivers. The support of multipoint communication in asynchronous transfer mode (ATM) networks is essential for applications such as audio/video conferences, distributed interactive simulations and replicated database synchronization. In recent years, issues regarding how to efficiently support multipoint-to-multipoint (mp–mp) available bit rate (ABR) service [3] have been studied extensively. A straightforward way to provide this service is combining the point-to-multipoint (p–mp) and the multipoint-to-point (mp–p) ABR services [21]. While many studies have been done on p–mp ABR service [9], [12], [13], [15], [17], [19], little literature is available on mp–p ABR service.

A switch algorithm, which supports mp–p ABR service can be functionally divided into two parts: rate allocation algorithm and merging algorithm. The function of the former is to calculate a proper cell rate for a connection to utilize the network efficiently and fairly. Several existing rate allocation algorithms for mp–p ABR services [7], [16], [20] have been developed by modifying point-to-point (p–p) rate allocation algorithms. In mp–p ABR service, cells from different sources are merged into the same virtual connection (VC) in some merge points. Therefore, as the resource management (RM) cells belonging to different VCs leave a merge point, they are indistinguishable to the merge point when they return. A merge algorithm has the responsibility to send backward RM (BRM) cells to appropriate upstream branches when a BRM cell is received. Furthermore, it has to maintain the ratio of the number of FRM cells sent by a source to the number of received BRM cells equal to one.

In most existing merging algorithms [7], [20], [21], a merge point forwards all the arriving FRM cells to the outgoing link. Since these FRM cells originated from different sources, the current cell rates (CCR), which indicate the sending rates of the sources, in the FRM cells may be different. As mentioned before, these FRM cells are indistinguishable when they leave the merging point, and the downstream switches will use the CCRs to calculate the proper cell rate for this VC. If a small CCR is used, the calculated result will be a small value. This will cause unnecessary rate reductions and link under-utilization. Furthermore, it is not essential to forward all arriving FRM cells and this will be explained in Section 4.

In this paper, we propose a merging algorithm, which is able to lessen the unnecessary rate reductions. As a result, it can achieve higher link utilization than existing algorithms. The proposed algorithm also enhances the function of a merging algorithm that has not been addressed in the literature. In existing studies, the action of merging refers to the ATM layer behavior that maps multiple VCs into a single VC. The proposed algorithm, however, really “merges” the FRM cells belonging to different VCs. In other words, fewer FRM cells are forwarded to the outgoing link. Although in our algorithm some FRM cells are merged, we still keep the basic function of a merging algorithm. That is, the ratio of the number of FRM cells sent by a source to the number of received BRM cells is equal to one. This guarantees that our algorithm does not cause the source to receive less network status carried by BRM cells. Another advantage of the proposed algorithm is that more data cells can be sent into the network because the overhead of control cells (i.e. FRM cells) is reduced.

The rest of this paper is organized as follows. In Section 2, we discuss the related issues for mp–p ABR service. Section 3 reviews two existing merging algorithms. In Section 4, we first state the motivation of this work and then describe the proposed merging algorithm. Section 5 addresses the numerical analysis of our algorithm. The analysis focuses on how much control overhead can be reduced for different network topologies. Simulation results are presented in Section 6 to evaluate the performance of the proposed algorithm. Finally, Section 7 concludes the paper and addresses the future research directions.

Section snippets

The cell interleaving problem

In ATM networks, a switch uses the virtual path identifier (VPI) and the virtual channel identifier (VCI) carried in a cell header to forward the cell. A source sends cells with the assigned VCI and VPI, which are obtained when the connection was established, to its adjacent switch. When the switch receives a cell, it uses the VPI/VCI to lookup which output port the cell should be forwarded to, and modifies the VPI/VCI. Eventually, the cell will arrive the destination via one or more switches.

Previous works

In this section, we review two existing merging algorithms. The first algorithm was proposed by Ren et al. in Ref. [20]. In their algorithm, once a merge point receives an FRM cell from an upstream branch, it will execute the rate allocation algorithm and return a BRM cell to the upstream branch immediately. In addition, it will forward the FRM cell to the outgoing link. When the merge point receives a BRM cell, it will retrieve the required information such as ER, CI and NI, and then discard

Motivation

Although Ren's algorithm is better than the first algorithm we mentioned, it has the following two drawbacks. First, it forwards all the arriving FRM cells to the downstream network. These FRM cells from different VCs are indistinguishable to the downstream switches when they leave the merge point. Most rate allocation algorithms, including ERICA, use the CCR field in FRM cells to calculate the allowed cell rate for the VC. Since the CCRs of different VCs are different, if a small CCR is used

Numerical analysis

In this section, we analyze the quantity of RM cells that can be reduced by our proposed algorithm. First, the analysis is based on a single merge point. Then we can use the analysis result to calculate the total number of reduced RM cells for an mp–p connection. We also discuss the performance of our algorithm under different network topologies. Especially, we are interested in the network topologies in which our algorithm will have either the best or the worst performance. Since it is hard to

Simulation results

To evaluate the performance of the proposed algorithm, we present the simulation results in this section. The simulation configuration used is shown in Fig. 6, in which each link has a capacity of 150 Mbps. The length of each link connecting adjacent switches is shown in the figure. The other access links are assumed to be 100 m long. There is a mp–p connection, which has four sources, src1, src2, src3 and src4, sending data to the same destination. In our simulation, we assume that the sources

Conclusions and future work

In this paper, we have proposed a merging algorithm, which has rarely been addressed in the literature. By forwarding the FRM cells belonging to the VC with the largest FRM-cell arrival rate, the proposed algorithms are able to decrease the unnecessary rate reductions. As a result, link utilization can be improved as compared to existing algorithms. To deal with the arrival of an FRM cell or a BRM cell, the running time of the proposed algorithm is O(1), which means that our algorithm does not

Acknowledgements

The authors would like to thank the anonymous reviewers who provided helpful feedback on the manuscript.

References (23)

  • W. Ren et al.

    Multipoint-to-multipoint ABR service in ATM

    Computer Networks ISDN Systems

    (1998)
  • Y. Afek et al.

    Phantom: a simple and effective flow control scheme

    Proc. ACM SIGCOMM

    (1996)
  • A. Arulambalam et al.

    Allocating fair rates for available bit rate service in ATM networks

    IEEE Commun. Mag.

    (1996)
  • The ATM Forum Technical Committee. ATM Forum traffic management specification version 4.1, ATM Forum AF-TM-0121.000,...
  • A. Charny et al.

    Congestion control with explicit rate indication

    Proc. Int. Conf. Commun.

    (1995)
  • Y.C. Chen et al.

    On the effective traffic control of ABR services in ATM networks

    IEICE Trans. Commun.

    (1998)
  • F.M. Chiussi et al.

    Explicit rate ABR scheme using traffic load as congestion indicator

    Proc. 6th Int. Conf. Comp. Commun. Networks

    (1997)
  • S. Fahmy, R. Jain, R. Goyal, B. Vandalore, A switch algorithm for ABR multipoint-to-point connections, ATM Forum...
  • S. Fahmy, R. Jain, R. Goyal, B. Vandalore, Fairness for ABR multipoint-to-point connections, Proc. SPIE Symp. Voice...
  • S. Fahmy et al.

    Feedback consolidation algorithms for ABR point-to-multipoint connections in ATM networks

    Proc. IEEE INFOCOM

    (1998)
  • J.M. Jaffe

    Bottleneck flow control

    IEEE Trans. Commun.

    (1980)
  • Cited by (0)

    View full text