38 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS VOL.6, NO.1 February 2008 Non-Flooding Bridging Solutions for Resilient Packet Rings Pisai Setthawong1 and Surat Tanterdtid2, Non-members ABSTRACT classes with di®erent service guarantees for support- Resilient Packet Ring (RPR) is one of the IP- ing various tra±c types. Bandwidth can be reserved based technologies that have been proposed to replace for time-sensitive tra±c such as VoIP and videocon- SONET/SDH in metropolitan area networks because ferencing. For best-e®ort tra±c such as web and ¯le it is well-adapted to handle diverse tra±c, includ- transfer, fairness algorithms ensure that the nodes ing multimedia tra±c, found in present-day networks. are dynamically allocated their fair share of the avail- Additionally, the RPR network is used e±ciently able bandwidth. The RPR network is also e±cient. with nodes transmitting simultaneously as long as Both dual-rings are used for packet transmission. The the paths of the packets do not overlap. However, packets are removed at the destination, allowing the when bridging RPR networks, packets are flooded on nodes to transmit simultaneously as long as the paths the bridged RPR network if the packet destination is of the packets do not overlap. on a remote network other than the source network. Although MANs can be implemented as a single As a result, the network is used ine±ciently, decreas- large RPR network spanning hundreds of kilometers, ing the available bandwidth for other tra±c. In this most carrier networks consist of multiple small-sized paper, we propose enhanced topology discovery pro- MANs, or access MANs, interconnected by inter-area tocol and enhanced spanning tree protocol to prevent MANs. This network architecture can also be im- the flooding of packets on the bridged RPR network. plemented using RPR. Each of the MANs is imple- Simulations were performed in order to evaluate the mented as a single RPR network and the RPR net- proposed solutions. The results show that the flood- works are interconnected by bridges. ing of packets is successfully prevented and that the However, the bridged RPR network is ine±cient. network is used more e±ciently as compared to the Packets are flooded on the bridged RPR network if RPR the packet destination is on a remote network other than the source network. The flooding decreases the Keywords: Resilient Packet Rings, IEEE 802.17 available bandwidth in the network for other tra±c. Standard, Bridging, Topology Discovery, Spanning Network e±ciency, a key bene¯t of RPR, is no longer Tree Protocol, Tra±c Engineering achievable. In order to achieve network e±ciency in bridged RPR networks, this paper proposes two so- 1. INTRODUCTION lutions to prevent the flooding of packets { enhanced topology discovery protocol and enhanced spanning The prevalent technology found in present-day tree protocol. metropolitan area networks (MANs) is based on The remainder of this paper is organized as fol- SONET/SDH. However, SONET/SDH is a circuit- lows. Section 2 presents background information switching technology that is ill-adapted to handle the about RPR. Section 3 describes previous research on bursty nature of multimedia tra±c increasingly found bridged RPR networks. In Section 4, we present the in present-day networks. IP-based networks, which proposed solutions. The proposed solutions are then bene¯t from statistical multiplexing, are more ap- evaluated by means of discrete event simulations in propriate for handling multimedia tra±c. Resilient Section 5. Finally, Section 6 concludes the paper. Packet Ring (RPR), also known as the IEEE 802.17 Standard [3], is one such technology. RPR is a dual- ring network that de¯nes a SONET/SDH reconcilia- 2. BACKGROUND tion sublayer. This sublayer enables RPR to operate The structure of an RPR network consisting of two on the same physical layer as SONET/SDH and to unidirectional ringlets and four nodes is shown in Fig- utilize the existing SONET/SDH physical infrastruc- ure 1. Both ringlets are used for data transmission, ture. with the tra±c on one ringlet flowing in a clockwise There are other key bene¯ts of using RPR instead direction (ringlet 0) and that on the other flowing in a of SONET/SDH. The former de¯nes three service counterclockwise direction (ringlet 1). In order to use these ringlets e±ciently, the nodes transmit packets Manuscript received on June 21, 2007 ; revised on October on the ringlet with the smaller hop count to the desti- 13, 2007. 1;2 The authors are with department of Telecommunications nation. For example, while node 1 transmits packets Science, Assumption University, Thailand, E-mail: to node 2 on ringlet 0, it transmits packets to node 4 Non-Flooding Bridging Solutions for Resilient Packet Rings 39 Node 2 Bridge 1 RPR 2 Bridge 4 Ringlet 0 Bridge 2 RPR 4 RPR RPR 1 Node 1 Node 3 RPR 3 Ringlet 1 Node 4 Fig.1: An RPR network consisting of two unidirec- Fig.2: A bridged RPR network consisting of four tional, counter-rotating ringlets and four nodes. RPR networks interconnected by three bridges. on ringlet 1 [2]-[9]. gether. Second, each network may be under di®erent administrative domains but interconnections between 2.1 Topology Discovery Protocol the domains are still desired, such as between di®er- ent faculties in the same university. Third, for carrier In order to determine the ringlet to transmit the networks, the RPR networks can be used to imple- packet on, the node maintains a topology database. ment access MANs that are bridged together to form The topology database includes a list of nodes on the inter-area MANs. network and the hop count to those nodes on each ringlet. The node maintains the topology database using 2.3 Spanning Tree Protocol the topology discovery protocol. The node broad- One of the major issues with bridging, not only in casts a topology discovery packet on both ringlets. RPR but also in other IEEE 802 networks, is the risk When the other nodes on the network receive the of creating loops within the bridged network. If a loop topology discovery packet, they learn about the ex- exists, there is a possibility that a bridge will continue istence of the node that originated the packet. The to copy packets to its connected networks without hop count between the nodes is then calculated as the realizing that the packets have already been copied to di®erence between the ttlbase and the ttl ¯elds in the those networks previously. The packets will continue packet. The ttlbase ¯eld contains the original time- to loop in the bridged network inde¯nitely and use to-live value of the packet and the ttl ¯eld contains bandwidth unnecessarily. In Figure 3, networks 1, 2, the current time-to-live value. and 3 form such a loop. Once all the nodes on the network have broad- casted their topology discovery packets, the nodes Bridge 4 will have a topology database of the network. The Bridge 1 RPR 2 node will then be able to determine the ringlet to Bridge 2 RPR 4 transmit the packet on with the smaller hop count to RPR 1 the destination. The node must continue to broad- cast the topology discovery packet periodically. If the node fails to do this, the other nodes on the network RPR 3 Bridge 3 will assume that the node has been detached and they will rebuild their topology database to exclude this node. Fig.3: Bridges 1, 2 and 3 form a loop consisting of RPR networks 1, 2, and 3. 2.2 Bridging with RPR The loops may be created unintentionally, espe- Multiple RPR networks can be bridged together to cially when the networks are bridged over multiple ad- form a bridged network, as shown in Figure 2. ministrative domains. On the other hand, the loops The specialized node that interconnects the RPR may also be created intentionally, for example, to pro- networks is known as the bridge. The bridge forwards vide redundant paths in the case of link failure. When or copies the packets that transit it on one network a link fails, an alternative path still exists between to the other network(s) it is connected to, in order to nodes. ensure that the packets reach their destination. In any case, if there are loops in the bridged net- There are many possible reasons for bridging mul- work, they must be removed. For this purpose, the tiple networks together. First, a single network can IEEE standard speci¯es the spanning tree protocol [?, support a maximum of 255 nodes and is optimized for ]. The result of executing this protocol is that par- a maximum ring length of 2000 km. When a greater ticular links to the bridges are identi¯ed to be deac- number of nodes or a longer ring length is required, tivated, removing the loops in the bridged network. the solution is to bridge multiple RPR networks to- This protocol involves the following steps: 40 ECTI TRANSACTIONS ON ELECTRICAL ENG., ELECTRONICS, AND COMMUNICATIONS VOL.6, NO.1 February 2008 ² Bridges periodically exchange con¯guration mes- 3.1 Bridging in the IEEE 802.17 Standard sages, also known as bridge protocol data units (BP- Bridging in the IEEE 802.17 Standard, hereafter DUs), among themselves. The BPDUs include: (1) referred to as standard bridging, makes the assump- the address of the bridge that created the BPDU, (2) tion that each node only knows the location of the the address of what the bridge has identi¯ed as the nodes that are on the same network as itself.