Network Convergence Over MPLS
Total Page:16
File Type:pdf, Size:1020Kb
Network Convergence over MPLS Enrique Vázquez, Manuel Álvarez-Campana, Ana B. García Dept. of Telematics, Technical University of Madrid, ETS Ing. Telecomunicación, Ciudad Universitaria, 28040 Madrid, Spain {enrique, mac, abgarcia}@dit.upm.es http://www.dit.upm.es Abstract. Multiprotocol Label Switching (MPLS) is emerging as a flexible technology that can transport voice, IPv4, IPv6, layer 2 services (Frame Relay, ATM, Ethernet, etc.), and even PDH and SDH/SONET circuits over a single packet infrastructure, in a new attempt to solve the old problem of network convergence. MPLS traffic engineering, quality of service support (in combina- tion with DiffServ), and fast restoration capabilities can be used to provide each service with strict service-level agreements (SLAs) in a cost-efficient way. Sev- eral standardization and industry organizations are contributing to this goal. This article reviews current activities in the IETF, MPLS/Frame Relay Alliance, and ITU-T related to transport over MPLS, comparing their approaches, identi- fying overlaps, and open issues. 1 Introduction About ten years ago, the Internet Engineering Task Force (IETF) and the ATM Forum specified several solutions to transport IP packets over Asynchronous Transfer Mode (ATM) networks, including Classical IP over ATM, LAN Emulation over ATM, and Multiprotocol over ATM. All of these followed a network overlay model where IP was put over ATM, and each of them retained their own control procedures. Concur- rently, several vendors proposed alternative solutions that sought a tighter integration of IP and ATM in some cases, or simply to forward packets at very high speeds based on fixed-length labels à la ATM, but without cells. For example, in IP Switching [1], promoted by Ipsilon Networks, IP packets are segmented into cells, and the ATM virtual connection identifier included in every cell header is used as the packet label. The standard ATM signaling and routing protocols are replaced by the label assignment protocols defined by IP Switching. In more gen- eral proposals, packet labeling is defined independently of the lower layers, so that la- beled packets may be transported over ATM or any other technology. See for exam- ple Tag Switching, promoted by Cisco [2]. The Multiprotocol Label Switching (MPLS) architecture, standardized by the IETF in 2001 [3], follows this general approach. Packets labeled with MPLS can be encap- sulated over different layer 2 protocols, including Ethernet, Point-to-Point Protocol (PPP), Frame Relay (FR), and ATM. MPLS label values may be copied to suitable fields in the layer 2 header, for example the Data Link Connection Identifier (DLCI) in Frame Relay, or the Virtual Path/Channel Identifiers (VPI/VCI) in ATM, so that labeled packets can be forwarded directly by FR or ATM switches with the appropri- ate control software upgrades. The initial focus on using labels to simplify packet forwarding and increase router performance quickly shifted to using labels as a powerful tool to tag selected packet flows and control the routes they follow across the network. In particular, MPLS is very flexible in terms of: − What is labeled: IP packets or other protocol data units. MPLS can transport voice and data, with or without IP. See the following sections. − How many labels are used: several labels can be applied to the same packet form- ing a label stack. Packets are classified and labeled when they enter the MPLS network. More labels may be added to the stack at convenient points inside the network. For example, if the link between two MPLS nodes A and B fails, node A may add a new label to packets in transit in order to forward them to B via an al- ternative path that bypasses the failure. When the packets arrive to node B this la- bel is removed. − What is the label meaning: for example, labels may be used to determine the packet route, to indicate a class of service, as a multiplexing identifier to distin- guish several flows that follow the same route, to separate packets corresponding to different virtual private networks, etc. − The method used to assign and distribute labels in the network: label values may be configured manually by the network administrator, or may be assigned automati- cally in different ways, for example coupled with the normal routing procedures in the network. Alternatively, labels may be assigned by exchanging signaling mes- sages from node to node along a particular route, similarly to a connection estab- lishment in an ATM network. The chosen route may have been selected by a con- straint-based routing algorithm taking into account the capacity required to serve a particular traffic flow. The MPLS capabilities are exploited in several important applications, including the implementation of IP Virtual Private Networks (VPNs) [4], Traffic Engineering (MPLS-TE) [5], support of Differentiated Services (DiffServ) [6], DiffServ-aware Traffic Engineering (DS-TE) [7,8], and fast rerouting [9,10]. Many aspects of MPLS are still subject of research and standardization. For example, at the time of writing this paper there were over 100 Internet Drafts related to MPLS. The focus has also shifted from transporting MPLS over different layer 2 technolo- gies to the opposite, i.e. transporting layer 2 protocols over MPLS. In particular, MPLS is well suited to implement Virtual Private LAN Services (VPLS) that extend Ethernet over the metropolitan and wide area. This article reviews the standards and implementation agreements for voice, layer 2, and layer 1 transport over MPLS produced by the IETF, the ITU-T, the MPLS/Frame Relay Alliance (MFA), and the ATM Forum. See Fig. 1. Section 2 compares the solutions proposed by ITU-T and the MFA for direct voice transport over MPLS, i.e. without encapsulating it in ATM cells or IP packets. Sec- tions 3 and 4 present the specifications for layer 2 and layer 1 transport over MPLS under development by the IETF and the MFA. ATM Forum specifications for ATM- MPLS interworking are also considered. Finally, section 5 outlines the evolution of MPLS-based applications. Ethernet, SONET voice voice IP ATM,… SDH PDH MFA IETF PWE3 MFA Martini AAL2 (IA5.0) VoMPLS (IA1.0) Drafts CEP AAL1 (IA4.0) MPLS CEP: Circuit Emulation over Packet MFA: MPLS/Frame Relay Alliance PWE3: Pseudo Wire Emulation Edge to Edge VoMPLS: Voice over MPLS Fig. 1. Network convergence over MPLS 2 Voice Transport over MPLS Voice packets can be transported over MPLS without the overhead associated to the typical RTP/UDP/IP encapsulation [11]. When several voice communications are transported between the same end-points, e.g. two voice gateways in Fig. 2, concate- nating voice packets from different communications before labeling and transmission helps to reduce the encapsulation overhead. In voice over IP, the RTP/UDP/IP headers may be compressed using different al- gorithms like IP Header Compression, Compressed RTP, Enhanced Compressed RTP, Robust Header Compression, and others defined in RFCs or Internet drafts. Concatenation may be implemented in different protocol layers. For example, Com- posite IP (CIP) and Lightweight IP Encapsulation (LIPE) concatenate voice packets above IP, while Point-to-Point Protocol Multiplexing (PPPmux) implements concate- nation at layer 2. Gateway MPLS network PSTN Voice network ISDN … LSR VoFR PSTN VoATM … Voice network Gateway Label Switching Router Label Switched Path Fig. 2. Voice over MPLS reference architecture In voice over MPLS, two main solutions have been proposed for concatenation. The first one, defined in the MPLS/FR Alliance (MFA) implementation agreement 1.0 [12], supports transport of multiplexed voice channels, various voice compression algorithms, silence removal and silence insertion descriptors, transfer of dialed digits, and channel associated signaling. Each concatenated voice packet is preceded by a 4- octet header that includes a channel identifier, a payload type, a counter, and a pay- load length field. See Fig. 3. If the payload length is not a multiple of 4 octets, up to 3 pad octets are included to make it word (32 bits) aligned. Up to 248 calls can be mul- tiplexed within a single Label Switched Path (LSP) identified by the outer MPLS la- bel. As an implementation option, additional inner LSPs may be created using stacked labels. 8 8 8 6 2 bits CID Channel Identifier CID PT C L P PT Payload Type C Counter VoMPLS header L Length in words (32 bits) P PAD length in octets payload payload payload Layer 2 1 or more MPLS labels length must be a Layer 2 overhead (4 octets each) multiple of 4 octets overhead Fig. 3. Voice over MPLS (MPLS/Frame Relay Alliance IA 1.0) The second solution addresses similar functions, but instead of defining a new voice encapsulation like [12], it reuses components of the ATM Adaptation Layer type 2 (AAL2), defined for transport of several variable bit rate voice and data streams multiplexed over an ATM connection. See ITU-T recommendation Y.1261 [13] and the new MFA implementation agreement 5.0 [14]. AAL2/MPLS is concep- tually similar to AAL2/ATM but replaces ATM connections with MPLS LSPs, so the ATM cell header overhead is eliminated. AAL2 may also be used over IP as dis- cussed in [15]. RTP AAL (1, 2, or 5) UDP VoMPLS AAL2 ATM IP (MFA IA 1.0) (MFA IA 5.0) MPLS Fig. 4. Protocol stacks for voice transport over MPLS 3 Layer 2 Transport over MPLS The transport of voice over AAL/ATM/MPLS shown on the left column in Fig. 4 is a particular example of transport of ATM cells, containing voice or any other type of information, over an MPLS network. The ATM Forum specifications for ATM- MPLS network interworking [16,17] define procedures to send a single ATM cell, a group of cells, or an AAL5 protocol data unit (PDU) encapsulated in a MPLS frame.