Voice Transport Techniques over MPLS

Voice Transport Techniques over MPLS Junaid Ahmed Zubairi, Ph.D., Member, IEEE with DiffServ for prioritized handling of the QoS traffic. ITUT has defined T-MPLS as a simplified version of MPLS for carrier transport networks. T-MPLS and its unique placement in the telecommunications industry are discussed briefly. Abstract- In this paper, we discuss currently evolving technologies for packet based real-time voice communication. Packet based voice has gained popularity in the toll-quality market due to its flexibility and cost effectiveness. IMS (IP Multimedia Subsystem) is being developed for converged voice Next, therotl a as evolved through ardsfr V oM agreedic ss carrier implementation agreements facilitated by IP/MPLS Forum [2]. and data services. While initially it was intended for mobile networks, it has been expanded to include support for fixed networks. In the wired network, the quality of service needed for media traffic can be provided by MPLS and related protocols. The traffic engineering and quality of service mechanisms of MPLS and Diffserv allow the differential treatment of premium traffic on predefined LSPs (Label Switched Path). Recently TMPLS has been introduced as the transport network for connection oriented packet switched traffic, dropping some of the MPLS features that were irrelevant for connection oriented applications. We introduce IMS and then look at implementation agreements, methods and protocols for transport of voice over MPLS (Multi Protocol Label Switching) networks. II. IP MULTIMEDIA SUBSYSTEM IP Multimedia Subsystem is SIP (Session Initiation Protocol) and IP based next generation architecture for unifying the data and media services for mobile and fixed network users. IMS is being developed by 3GPP, a liaison body that works with other organizations including TISPAN (Telecom and Intemet converged Services and protocols for Advanced Networks). Index Terms- IP multimedia subsystem, VoMPLS, T-MPLS, Quad Play, QoS, jitter,, A2oMPLS IMS-S I. INTRODUCTION oice over IP has been around for several years. Users [have experimented with various audio-conferencing tools offered by messaging services. Discounted phone cards that route calls through the Internet have been sold for quite some time. However, the end to end VoIP telephony did not achieve notable success until recently. The number of worldwide VoIP customers has now reached 40 million and it is projected to grow to approximately 250 million by the end of 2011 [1]. The quad play (voice, video, wireless and data) services are being defined for next generation integrated networks in IMS (IP Multimedia Subsystem). This tutorial paper discusses the IMS specifications and focuses on the techniques and methods developed for transport of voice over MPLS. The protocols are developed by 3GPP (3rd Generation Partnership Project), IETF (Internet Engineering Task Force), IP/MPLS Forum (formerly MFA Forum / MPLS Forum) and the ITU-T (Telecommunication Standardization Sector of the International Telecommunication Union). Figure 1: IMS Architecture IMS goals include provisioning QoS, enabling network usage billing for media applications and integrating various media services. IMS adds a control layer below the services and defines interfaces to the mobile and landline phone network as well as the IP network. As shown in Figure 1, IMS In the next section, the key proposals of IMS are defined. MPLS serves as an enabling technology for QoS, and it works Manuscript received November 2008. Junaid Ahmed Zubairi is with the Department of Computer Science, State University of New York, Fredonia, NY 14063 USA ( phone: 716-6734694; fax: 716-673-4821; e-mail: Zubairi@fredonia.edu). defines CSCF (Call Session Control Function) server as the primary SIP server that has three functions. The Proxy CSCF ©2008 IEEE. Personal use of this material is permitted. However, 25 permission to reprint/republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE . The MPLS header fields include an unstructured 20-bit label that identifies the LSP to which this packet belongs. The label is assigned based on the FEC (forwarding equivalency class) of the traffic trunk. It is followed by 3 bits TC (Traffic Class) field and 1 bit indicating whether this is the bottom of the label stack. The TTL (Time to Live) field has the same usage as the TTL field in an IP header. Class based queuing (CBQ) is invoked on routers that implement the differentiated services (DiffServ) protocol to create separate queues for each class of traffic. DiffServ divides the traffic into expedited forwarding (EF) and assured forwarding (AF) per-hop behaviors (PHBs). EF is the premium service offered under DiffServ. It is suitable for low latency and low jitter flows that maintain almost constant rate. In an MPLS-DiffServ network, the routers jointly implement various MPLS and DiffServ functions. The ingress router is responsible for determining an LSP for a new flow request. The QoS requirements of the new flow can be translated into DiffServ class assignment at the ingress. For this purpose, the TC (Traffic Class) field in the MPLS shim header is used. As the TC field is 3 bits in length, it can represent only eight different scheduling and drop precedences. Under MPLSDiffServ [5], two types of LSPs are defined. E-LSP interprets the label field of MPLS shim header as the egress identifier and the TC field as the DiffServ PSC (PHB scheduling class) combined with the drop precedence. On the other hand, LLSP interprets the label field as the DiffServ scheduling priority and the destination. The TC field in L-LSP is used to indicate only the drop precedence of the packet. The main difference between E-LSP and L-LSP is the aggregation feature in E-LSP resulting in scalability; however, some PSCs in an E-LSP may suffer because the bandwidth is reserved for the whole LSP [6]. Figure 3 shows the placement of the DiffServ class identifier and MPLS label for an E-LSP. lets client user agents connect to the IMS , the Serving CSCF provides the basic signaling for IMS session and Interrogating CSCF bridges the different carriers thus enabling roaming. HSS (Home Subscriber Server), in association with SLF (Subscription Locator Function) , runs a database listing users and their subscribed services. HSS would be used to authenticate the users that request IMS session. IMS links to all types of networks including wired IP network, wireless CDMA/GPRS network and fixed phone network. PSTN is supported by MGCF (Media Gateway Control Function)[3]. IMS uses IETF standardized protocols in order to make it widely acceptable. However, the details of implementation are not specified, thus leaving it flexible. Vendors may come up with products that are IMS-compatible but the same may be implemented with different strategies. Only the interfaces with network may follow IMS standards. III. MPLS AND DIFFSERV PROTOCOLS Real-time voice traffic has tight timing budget with the required end to end delay values below 150ms and the jitter or delay variation around 50ms. The best effort protocols cannot guarantee such limits because the datagrams do not follow a fixed path and may arrive at the destination out of order. With increased congestion, the queues get longer resulting in increased jitter. These problems make conventional IP networks largely unsuitable for connection-oriented applications such as interactive real-time voice. The real-time audio-conferencing applications written for the conventional IP networks use forward error correction and inter-mixing of small packets to recover from packet loss, often leaving the perceived call quality as less than desirable. MPLS [4] has emerged as the key integration technology for carrying voice, video, and data traffic over the same network. In an MPLS-enabled network, LSPs (Label Switched Path) are installed from an ingress node to an egress node prior to start of transmission. Each LSP can be specified with features that include time constraints and reliability. Therefore, the connection-oriented applications can take advantage of the "virtual connections" set by MPLS that satisfy some constraints. Since the LSPs are stackable, traffic from different flows sharing some common characteristics can be aggregated on an LSP. These characteristics may include common egress and identical QoS and protection requirements. MPLS operates by defining a label inside MPLS "shim header" that is placed on the packet between the layer-2 and layer-3 headers. The 32-bit shim header is organized as shown in Figure 2. It is referred to as "shim" header as MPLS allows la yer-2 and layer-3 to fit together firmly. MPLS Ladbel (20 TC (3 | (1* TTL (8 | bits) I bit) bits) bits) MPLS LAbel (20 TC bits) (FEC (DiffSCrv Dti)) S (1 bit) TTL (8 bits) IV. TMPLS PROTOCOL ITU-T believes that the transport networks do not have highly varying traffic patterns thus there is no need for some of the features of MPLS that largely suit the dynamically changing traffic of Internet. They introduced T-MPLS in 2006 for support of connection oriented packet switching flows. The T-MPLS (Transport MPLS) (ITU-T G.8110.1) protocol implements a simplified MPLS architecture for carrier class networks. T-MPLS adds the TE capabilities of SONET/SDH to Layer 2 network designs [7]. It provides backup tunnels with 50ms protection switching, thus enabling the migration from SONET/SDH transport to a fully packet switched netw ork ith The Figure 2: MPLS header fields 26 com standard parable reliability[8]. documents approved were G.81 10.1 the MPLS bypassing the RTP/UDP/IP protocol stack. As seen in Figure 4, the VoMPLS protocol stack is more compact as compared to VoIP. The MPLS UNI (User Network Interface) definition [14] allows users at CE (Customer edge) to establish LSPs to the PE (Provider Edge) network. The LSP may be a single link or a layer-2 network but it is considered a single hop LSP. Since there may be thousands of CE-to-CE connections, a potential scalability issue arises. The MPLS proxy admission control [15] solves the scalability problem by letting the PE ingress router search for existing tunnels that can satisfy the traffic parameters TLV in the label request message sent by the CE. Once a suitable LSP is found, its RIL (Resource Index Label) is returned to the CE which then uses it to encapsulate the traffic for that tunnel. QoS is guaranteed only within the PE-to-PE network. For CE-to-PE links, customers have to deploy layer-2 QoS techniques for provisioning of resources. Architecture of T-MPLS layer network, G.8112 Interfaces for the T-MPLS hierarchy (TMH) and G.8121 Characteristics of T-MPLS equipment. T-MPLS is described as a connection oriented packet transport technology that uses the MPLS label swapping and forwarding, carrier networks performance monitoring, simplified protection switching and ASON/GMPLS control and management functions [9]. It involves setting up unidirectional and bidirectional P2P (Point to Point) connections that take identical network path in both directions and receive TE support and management. T-MPLS simplifies end to end OAM (Operation, Administration and Management) by removing the IP specific functionality. For example, T-MPLS removes the PHP (Penultimate Hop Popping) feature that would require IP processing of the packet at the end router and similarly the ECMP feature is removed in order to avoid possible source identification confusion. The T-MPLS tunnel can carry multiple L2 and L3 services including IP/MPLS LSPs and PWE3 pseudowires. Companies that deploy T-MPLS are expected to find it easier and economical in terms of man-hours of staff training, as compared with IP/MPLS networks. It is because of the fact that the T-MPLS network's architecture is similar to SONET/SDH network. Recently, in early 2008, ITU-T and IETF have deliberated to resolve inconsistencies between MPLS and T-MPLS and the consensus is emerging to define separate code points for MPLS and T-MPLS in order to avoid confusion. IETF and ITU-T agree that MPLS and T-MPLS are disjoint networks. An LSP initiated from either network would encapsulate into Ethernet before transiting the other network. Client support in T-MPLS is based on IETF Pseudo wire model for Layer-2 VPN. The two-layer architecture includes top layer as client Virtual Circuit LSPs and bottom layer for aggregating the VC LSPs into Trunk LSPs. The key differences between MPLS and T-MPLS include the use of bidirectional LSPs in T-MPLS and no PHP, no merging of LSPs and no ECMP routing in TMPLS [10]. o RT1PLS MA JA M ALIA MPL5 AL3 A IJDP Figure 4: VolPoMPLS and VoMPLS protocol stacks B. MPLS UNI With Proxy Admission Control In voice deployment over MPLS, the VGW (Voice Gateway) acts as the CE requesting guaranteed LSPs to the remote VGW. The MPLS proxy admission control allows the voice gateways to dynamically request LSPs to remote VGWs and to share the multiplexed LSP-TE tunnels among themselves through the local PE acting as a proxy for admission control [16]. Using voice gateway as an example as in Figure 5, the following exchange of control messages may occur between the VGW and PE before a call can be established: 1) Phone 1 requests a voice call to Phone2. 2) VGW sends a label request message to PE containing FEC (Forwarding Equivalence Class) and Traffic Parameters TLV for the intended call. 3) PE acts as a proxy for the local VGW and searches existing TE tunnels to remote VGW that satisfy the traffic parameters and FEC in the label request message 4) If an existing tunnel is found that can satisfy the request, PE sends its RIL to the requesting VGW 5) The requesting VGW can now initiate the call and then release the resources once the call is terminated. V. VOICE TRANSPORT OVER MPLS Voice is carried on MPLS network either directly or with VoIP stack i.e. RTP/UDP/IP/MPLS. On the other hand, VoMPLS [11,12] carries voice directly over MPLS, reducing the headers significantly. Figure 4 shows the comparison of protocol stacks in different schemes. VoIP has the clear advantage of providing end-to-end connectivity with mature protocols. The MPLS user-to-network interface (UNI) definition for bringing users in direct contact with MPLS networks has been defined recently [13,14,15]. Using MFA IA 1.0 and IA 5.0, voice can be directly encapsulated in MPLS packets. The other option is to use the ATM over MPLS definition to enable VoATM voice service. Among these choices, VoMPLS is the most promising solution as it involves the least amount of protocol tax. From the above, it is obvious that the MPLS UNI together with MPLS Proxy admission control provides a powerful mechanism to deploy thousands of voice calls through the PE network without control messaging overflow. A. VoMPLS In VoMPLS,-ktd the packets are directly transpo ed over lt voice r 27 C. VoMPLS Header Formats In VoMPLS, Various header formats are defined for different payload types. Payloads may include encoded audio information, dialed digits, silence descriptors and control signaling. Two types of frames are defined in VoMPLS IA 1.0: primary and control frames. The VoMPLS primary frame header is shown [10] in Figure 6. Many primary VoMPLS frames may be multiplexed within a single MPLS packet. Pb- Figure 5: MPLS UNI and Proxy Admission Control Channeil Wfidtifie (8 bits) Pay load Type Counte (8 bits) Payload Length (6 bits) Pad Length (2 bits) 1) Using MPLS signaling mechanism, a bidirectional LSP is created 2) As voice call request arrives at the edge of the MPLS network, an existing CID is allocated to the new call within the LSP just created or CID signaling is done to establish the new channel. 3) Optionally, inner LSPs may be created within the outer LSP in which case the outer LSP label, inner LSP label and CID unqul idntf th voc. al the voice call. uniquely identify The VoMPLS control frame header is also 4 bytes in length. Control fames cannot be multiplexed and must be carried separately however the mandatory outer label and the optional inner label precede the control frame header fields in order to uniquely identify the voice call for which this control frame is sent. In the control frame, the Payload Type field distinguishes Pbetween dialed digits and the channel related signaling. Time Stamp is relative to the first randomized time stamp. The Redundancy field is very important to ensure the receipt of the control frames. If the Redundancy field is set to 0, 1, or 2, the control packet is repeated that many times. Another solution for VoMPLS reuses ATM Adaptation Layer 2 (AAL2) components but replaces ATM by MPLS [6,1 1], thus eliminating the ATM cell overhead. The MFA 5.0 implementation agreement proposes voice trunking over MPLS by directly encapsulating AAL2 common part sublayer (CPS) packets into MPLS (A2oMPLS). The gateway to the MPLS network should be able to function as an AAL2 switch. Multiple A2oMPLS connections may be multiplexed into a single LSP. One MPLS frame may carry multiple CPS packets. The A2oMPLS sub-frames may be of different length but the maximum CPS packet payload length is restricted to between 45 octets to 64 octets. Like the MFA IAlO, the CID field in the CPS packet header allows up to 248 A2oMPLS connections to be multiplexed. Each A2oMPLS connection can be uniquely identified with outer label of the LSP, optional inner label and the CID value. When inner labels are bits) Figure 6: VoMPLS MFA 1 primary frame header fields VoMPLS primary frame header is 4 bytes long and includes 5 fields: > Channel Identifier uniquely identifies the voice channel that is the source of the payload. Thus, a total of 248 different voice calls can be multiplexed into a single LSP. Payload Type identifies the encoding scheme used as well as silence removal/insertion descriptors. A value equal to or above 224 indicates control payload (part of the control frame) that would allow DTMF (dual tone multi frequency) dialed digits as well as signaling for the channel to be carried [4]. > The revolving Counter field is set at the first sample or frame and keeps incrementing for each additional frame. > Payload Length is read in conjunction with the pad length to keep the payload a multiple of 4 bytes. Multiple primary frames may be multiplexed with the use of optional inner MPLS labels in addition to one mandatory outer MPLS label. Using CID, up to 248 voice calls can be aggregated. The payload of primary frame may consist of encoded audio data or SID (Silence Insertion Descriptor) parameters. Keeping the frames and header formats in perspective, the voice calls can be established as follows [4]. used the number of calls that are . carried by a single LSP r increases rapldly. The following fields are included in the A2oMPLS header: Reserved (10 bits) currently ignored. > Length (6 bits) used for padding length. It is set to 0 if A2oMPLS packet length exceeds 64 bytes. > Sequence Number (S No) (16 bits) used if guaranteed ordered packet delivery is required. Sequence number of 0 indicates otherwise. The following fields are inside the CPS packet header; several CPS packets may be packed in one MPLS frame. > CID (8 bits) identifies the A2oMPLS connection carried. Thus a total of 248 connections (8 to 255) can be multiplexed into a single LSP. > LI (6 bits) identifies the length of the CPS packet. > UUI (5 bits) is used for user-to-user indication (0 to 27 for users, 30-31 for layer management, and bv 1rsre) L 28 HEC (5 bits) (header error control) uses CRC checksum. However, this feldmaynobeusei field may not be used in cheksu.'Hweer,thi the MPLS environment. [8] Industry is moving towards integration of phone system with the packet switching network. IMS is being developed to provide uniform interface for users from cellular, landline or wired systems. In wired and wireless packet based networks, MPLS can work with other protocols to guarantee the bandwidth needed for a voice call and provide fast protection switching. Voice can be deployed over MPLS using a variety of techniques that have been developed recently. In this paper, these techniques including VoIPoMPLS and VoMPLS are [11] MFA. 2001. Voice over MPLS: Bearer Transport Implementation Agreement (MFA IA 1.0). MFA Forum. [12] MFA. 2003. Voice trunkingformat over MPLS. MPLS/FR 5.0.0. 1.366.2. MFA Forum. [13] ITU-T. 2002. Recommendation Y.1261. Service requirements and architecture for voice services over MPLS. [14] [14] MFA 2003A D. Sinicrope, A. Malis, MPLS PVC User to Network Interface, MPLS/FR Alliance 2.0.1, May 2003. [15] MFA 2004A MPLS Proxy Admission Control Protocol Implementation Agreement 7.0.0. [16] Fineberg V, 2004 (With Sinicrope D, Phelan T, Sherwin R and Garbin discussed. D) The MPLS UNI And End-to-End QOS. Business Communications VoIPoMPLS deploys voice over the RTP/UDP/IP protocol stack, which uses MPLS tunnels. Since the number of protocols involved iS large, the control information equals or exceeds the payload. Most efficient method is to run voice directly over MPLS. VoMPLS is discussed in detail including Review Dec 2004 Pages 27 - 32. [17] Fjellskal, E., and S. Solberg. 2002. Evaluation of Voice over MPLS (VoMPLS) compared to Voice over IP (VoIP) (Masters thesis, Agder Univ College). Junaid Ahmed Zubairi received his BE (Electrical Engineering) from NED University of Engineering, Pakistan and MS and Ph.D. (Computer Engineering) from Syracuse University, USA. He worked in Sir Syed various proposals for enabling it in the network. VoMPLS is well suited for multiplexing in the core and carrying the bulk of voice calls through the MPLS domain. The difference between MPLS and T-MPLS, a new ITU-T proposal for carrier networks, is highlighted. The MPLS UNI proposal presented in 2004 is outlined. Since the original proposal was not scalable, proxy admission University Pakistan and Intl' Islamic University Malaysia before joining SUNY at Fredonia in 1999 where currently he is an Associate Professor in the Department of Computer Science. Dr. Zubairi is a recipient of many awards including Malaysian Government IRPA research award, National Science Foundation MACS award, SUNY Scholarly Incentive award and SUNY Faculty Fellowship award. He has published several book chapters, journal articles and conference papers in his areas of interest. His research interests include network applications, traffic engineering and performance evaluation of networks. He can be reached at junaid.zubairi fredonia.edu. d control was introduced to pack new control LSPs intointrodu existing to tunnels. With the introduction of IMS, many technologies would be integrated to allow exciting new features in packet voice. For example, transparent connectivity through landline and mobile networks would let the user continue the conversation while changing the connection from landline network to mobile network or vice versa. Since IMS is based on IPV6, the IMS adoption would be slow in parts of the world where IPV4 still runs the network. The VoP phones would be flexible and allow all the functionalities of current PSTN phones. Security of packet based phones and authentication of users poses new challenges that must be analyzed and resolved for widespread deployment. REFERENCES [1] Teral, 2006 Service provider and enterprise IP telephony markets, Infonetics Research. 30 Aug 2006. accessed Nov 4 2008 [2] Zubairi, J. 2008. Emerging Methods for Voice Transport over MPLS. Chapter in Taylor Francis Handbook of MPLS Technologies Dec 2008 [3] M. Hunter et. al. Security Issues with the IP Multimedia Subsystem (IMS) , Proc. Middleware Conference 2007, ACM. [4] Rosen, E. et al. 2001. Multiprotocol Label SwitchingArchitecture. RFC [6] [7] 2008. Barry, D. 2007. T-MPLS and PBTIPBB-TE offer connection-oriented packet transport. Lightwave May 2007 http://lw.pennnet.com. [9] ITU-T 2006 Draft new Amendment 1 to G.81 10.1 Y.13 70.1, Nov 2006. [10] Lum, 2006 When Networks Collide: Putting the T into MPLS 2006 article FiberSystems.org online h accessed Nov 42008. VI. CONCLUSION [5] ffic-engineering-for-Ethernet:-PBT-vs-T-MPLS/ accessed Nov 4 3031. Faucheur and Lai IETF RFC 4125, Maximum Allocation Bandwidth Constraints Modelfor Diffserv-aware MPLS Traffic Engineering 2005. Fineberg, V. 2003. QoS Support in MPLS networks. MFA Forum white paper Lunk, P 2008 Traffic engineering for Ethernet: PBT vs. T-MPLS Online LightWave article, 29