Several Issues in the Development of Packet Transport Networks

This work was funded by the National High Technology Research and Development Program of China (“863” program) under Grant No. 2007AA01Z2A4.

    There is no standard definition for Packet Transport Networks (PTN). Broadly speaking, PTN can be defined as any transport network that is based on packet switching and meets certain Operation Administration and Maintenance (OAM), protection, and network management requirements. Specific packet-switched technologies include Multiprotocol Label Switching (MPLS), Transport MPLS/MPLS Transport Profile (T-MPLS/MPLS-TP), Ethernet, Backbone Bridge Traffic Engineering (PBB-TE), and Resilient Packet Ring (RPR). Over the past two years, both T-MPLS and PBB-TE have been viewed as the leading PTN technologies. However, as support for PBB-TE from vendors and operators wanes, China has come to view T-MPLS/MPLS-TP as PTN. Therefore, PTNs described in this paper are based on T-MPLS/MPLS-TP.

 
    Transition from T-MPLS to MPLS-TP reflects a history of competition and integration of the transport and data fields. MPLS-TP is an achievement preceded by years of competition and negotiation between the Internet Engineering Task Force (IETF) and International Telecommunication Union’s Telecommunication Standardization Sector (ITU-T). It can be seen as a balanced exploitation of the benefits of transport and data fields. IETF currently leads the development of MPLS-TP standards, while ITU-T SG15 has a lesser hand and mainly relies on the participation of enterprises and individual expert members[1-6].

1 QoS in PTN
QoS provides predictable service quality in terms of packet loss ratio, delay, jitter, and bandwidth during network communication. QoS in PTN includes stream classification, labeling, rate limitation, bandwidth guarantee, traffic shaping, and scheduling strategy. MPLS Traffic Engineering (TE) and Differentiated Service (DiffServ) mechanisms are used to implement QoS in PTN, with the goal of creating a service-oriented end-to-end QoS guarantee[7-10].

1.1TE
According to IETF’s definition, MPLS-TP must support TE so that network resources can be controlled. The purpose of TE is to optimize resource use for effective and reliable network operation. An important element of TE is the Constraint-Based Routing (CBR) mechanism. In an
IP/MPLS network, TE is generally implemented by MPLS-TE, and has two important roles:

    (1) Making Service Routing Controllable

    A PTN service is first encapsulated by the Pseudo Wire (PW), and then multiplexed to the Label Switched Path (LSP). The LSP is established through network management or the control plane, and LSP routing implemented by either of these is controllable.

    (2) Making Service Bandwidth Controllable

    PTN can bear E1 emulation services and Ethernet services. The bandwidth of an E1 emulation service must be fixed and controllable, having high priority, no packet loss, and low delay. Ethernet services can be divided into Constant Bit Rate (CBR) services and Variable Bit Rate (VBR) services. A CBR has the same basic requirements and controllability as E1 emulation services. A VBR service uses Committed Information Rate (CIR) and Excess Information Rate (EIR) to control bandwidth. An operator only guarantees user bandwidth not exceeding the CIR, and may discard EIR traffic when the network is congested. In this way, network bandwidth is controlled.

    More specifically, the operator first configures the CIR of the PW (eg. the service) as well as the LSP, and sets the conditions for controllable connection. The sum of all the PW CIRs in one LSP should not exceed the CIR of the LSP; the sum of all the LSP CIRs in one link should not exceed the CIR of the link. Under these conditions, the network can satisfy the CIR bandwidth requirements of all services. Since VBR service involves bursts (eg. the EIR part), network congestion cannot be completely avoided using TE. DiffServ is an effective solution for ensuring CIR bandwidth in such conditions.

1.2 DiffServ
DiffServ is derived from Integrated Service (IntServ), and aims to provide differentiated service levels for Internet traffic. Compared with IntServ, DiffServ has a simpler control system with coarser granularity. It is used to control each QoS type after streaming convergence. IntServ, on the other hand, aims to control each individual stream. DiffServ is therefore scalable, and capable of QoS in large-scale networks.

    DiffServ classifies IP streams into different types at the edge of its domain, and assigns one Differentiated Services Code Point (DSCP) to each stream type. The core router in the domain checks the DSCP value, and dispatches packet forwarding according to different types of Per-Hop Behavior (PHB). IETF has defined two kinds of PHB: Expediated Forwarding (EF) and Assured Forwarding (AF).

    (1) EF
    EF PHB traffic is unaffected by any other PHB traffic, and this ensures packets are forwarded at the quickest possible rate. Similar to legacy leased lines, EF PHB can guarantee bandwidth service with low packet loss, low delay, and low jitter. CIR is the only bandwidth parameter of an EF-based service; EIR is equal to 0 and traffic greater than CIR is discarded. EF can be applied to E1 emulation and CBR Ethernet services, and should follow RFC3246.

    (2) AF
    AF provides four levels of packet forwarding, each level with three discarding priorities.  The level of a service is determined by PTN equipment—which configures forwarding resources (such as buffer and bandwidth) at different levels, and accounts for the discarding priority. If a service is not congested, AF service performances at different levels are equal. However, if the service is congested, packet loss occurs on all AF levels, and the degree of packet loss correlates with the service level. AF should follow RFC2597. 

1.3 DiffServ Supported by MPLS
MPLS-TP based PTN must use the MPLS DiffServ mechanism defined in RFC3270.

    After an IP packet has been encapsulated through MPLS, the core router cannot locate DSCP. Accordingly, IETF has proposed a DiffServ-supported MPLS that can map multiple Behavior Aggregates (BA) of DiffServ to a MPLS LSP and forward the traffic on the LSP according to the BA PHB. There are two mapping modes between LSP and BA: EXP-inferred-PSC LSP (E-LSP) and Label-Only-Inferred-PSC LSP (L-LSP).

    (1) E-LSP
    E-LSP uses the MPLS-labeled EXP field to designate multiple BAs to one LSP. The MPLS-labeled EXP field represents the PHB of a packet. Up to eight BAs can be mapped into the EXP field; that is, an E-LSP supports a maximum of eight service levels.

    (2) L-LSP
    L-LSP designates one LSP to one BA, and uses EXP to represent the packet discarding priority. An L-LSP can only support one service level.

    MPLS equipment switches label values per hop, but the management of mapping between label and PHB is difficult. Therefore, compared to L-LSP, E-LSP is more easily controlled because it can determine in advance the mapping relationship between the EXP field and PHB for every packet in the network. PTN equipment currently uses E-LSP. 

1.4 End-to-End QoS Implementation in PTN
TE and DiffServ supported by MPLS help PTN guarantee service-oriented end-to-end QoS. MPLS TE is used to control service routing and bandwidth in order to avoid network congestion caused by unbalanced loading. Once the network is congested by burst service or network protection, DiffServ supported by MPLS is used to guarantee CIR.

    Table 1 classifies service levels for E-LSP. Peak Information Rate (PIR)=CIR+EIR; and under this condition, the EXP value of LSP and PW in a data frame are the same one.

    Both E1 emulation services and CBR Ethernet services (such as voice and video) adopt EF PHB with a setting of CIR=PIR.

    Burst-type services such as virtual private networks and private Ethernet lines use AF PHB. To ensure CIR bandwidth of burst-type services, service streams must be measured, shaped, and labeled at the network ingress according to bandwidth parameters, and Two Rate Three Color Marker (trTCM) must be supported. Moreover, the EXP value of the data frame is set from the mapping relationship, which is used by the LSP follow-up nodes to select suitable PHB.
As for ordinary data services, CIR is set at 0, PIR is set, and Default Forwarding (DF) is used.

    Even if the network is congested, the service bandwidth of both EF PHB and CIR part of AF PHB traffic is always guaranteed. Ordinary data services are either discarded or weighed with AF PHB traffic so that certain bandwidth can be obtained during the congestion.

2 Layered Architecture of PTN
IETF RFC5654 divides an MPLS-TP system into transport service layer, transport path layer, and section layer. The transport service layer can be PW or service LSP, similar to Synchronous Digital Hierarchy (SDH) Virtual Channel (VC)-12. PW is used to offer emulation services such as Time Division Multiplexing (TDM), Ethernet, and Asynchronous Transfer Mode (ATM) services. Service LSP is used to offer network-layer IP and MPLS services. The transport path layer refers to the LSP layer, similar to SDH VC-4. The section layer is used to converge information from the transport service layer and transport path layer between two adjacent MPLS-TP nodes. The section layer can be implemented by either MPLS-TP or technologies such as SDH, Ethernet, or Optical Transport Network (OTN). With a layered architecture, PTN can achieve scalability similar to SDH/OTN.

    Besides 3-layer MPLS-TP, PTN should also support related functions of the service layer and section layer. Such functions include OAM of the Ethernet service layer (specified in IEEE802.1ag and Y.1731), OAM of the Ethernet link layer (specified in IEEE 802.3ah), overhead handling of SDH services and links, and protection.

    Current PTN equipment uses PW to support various emulation services but does not support IP/MPLS services through LSP. IP/MPLS service implemented by Ethernet PW emulation is highly transparent, but also inefficient—especially for short packets because Ethernet frame headers need to be transported. IP/MPLS implemented by TDM PW emulation has high requirements on network performance, and may increase equipment cost. IP/MPLS implemented by service LSP can avoid the abovementioned problems, but has poor service transparency and is possibly required to handle part of the L3 protocol. Therefore, service transparency, transport efficiency, and cost must be taken into account when selecting IP/MPLS service implementation technology.
In addition, current PTN equipment only supports Single-Segment Pseudo Wires (SS-PW); that is, where the source & sink of PW coincides with that of LSP. SS-PW cannot converge PWs borne by multiple LSPs, and requires PTN equipment to have high LSP capacity. Moreover, current PTN equipment with only end-to-end LSP protection cannot cope with multi-point faults. Multi-Segment Pseudo Wires (MS-PW) can be introduced to solve problems caused by SS-PW, and thereby improve PTN scalability. IETF has listed MS-PW as optional for MPLS-TP.

3 The Support of PTN on L3 Functions and Services
Currently, PTN mostly offers L2 services, including E1/ATM emulation services and E-Line/E-LAN/E-Tree services. It is primarily applied in 3G and Long Term Evolution (LTE) mobile backhaul networks. PTN can satisfy the bearing requirements of 3G networks, but it is still doubtful whether it can meet the bearing requirements of future LTE systems.

    LTE systems have different bearing requirements from 3G networks because they require interconnection between base stations (X2 interface) and multi-homing from base station to Service Gateway (SGW). There are two ways to satisfy the bearing requirements of LTE: end-to-end router networking, and L3+L2 networking (with L3 networking for the core layer and L2 networking for the convergence and access layer). Since end-to-end router networking still has problems with network scalability, manageability, and controllability, L3+L2 networking has gained recognition and support. The core layer in this solution can be implemented by router networking or by introducing L3 functions into PTN. 

    L3 functions mainly include IP routing and forwarding, L3 MPLS Virtual Private Network (VPN), and L3 multicasting. IP traffic and multicast, with uncertain traffic bandwidth and routing, cannot provide strict QoS guarantee. If these two services are introduced into PTN, they should be set as the lowest level services in order to avoid any impact on L2 services. L3 MPLS VPN can use MPLS TE and DiffServ mechanisms to guarantee QoS. MPLS VPN can also support L3 multicasting with QoS guarantee.

    For PTN networks, L3 services with QoS requirements can be offered by L3 MPLS VPN, while L3 services with no QoS requirements may be implemented directly through IP routing and forwarding.

4 Data-Plane Loopback Functions
Current PTN equipment only supports OAM Loopback (LB). OAM LB can be used to verify bidirectional interconnection between the source and sink maintenance endpoints, and to check for inter-node/intra-node faults. However, it cannot locate the specific fault positions. As shown in Figure 1, if there is a fault on the PE2-PE3 link, OAM LB cannot determine whether the fault is in PE3 or on the link. If PTN equipment can support SDH-like data-plane LB; that is, service LB, accurate fault positioning can be implemented through LB of different points.
Similar to SDH, PTN data-plane loopbacks include Remote/Ingress Loopback, Local/Egress Loopback, and Fiber/Client Loopback. Beside for fault locating, fiber loopback can be used to perform single-end service performance tests on bidirectional time delay, packet loss ratio, and throughout. It also supports testing in live networks.

    Supporting both remote and local OAM loopbacks, PTN can implement functions similar to remote and local loopbacks on the data plane. Therefore, fiber loopback should be implemented in PTN to accurately locate faults, and for single-end testing. Further research is still needed to determine whether PTN should support data-plane remote and local loopback functions. Currently, IETF and ITU-T are discussing the standardization of data-plane loopback functions.

5 Conclusion
PTN is the optimal solution for evolving Multiservice Transport Platform (MSTP) networks bearing 2G mobile backhaul. The goal is to meet requirements of high-quality services such as 3G mobile backhaul and enterprise private line services. From 2008 to 2009, China’s three dominant telecom operators trialled PTN bearing 3G mobile backhaul in their live networks. They pushed forward the maturing and commercialization of PTN products. China Mobile began large-scale purchasing of MPLS-TP based PTN equipment in October 2009, marking a turning point in the industrialization of PTN.  International MPLS-TP standards have attracted much attention in 2010, and the stability of international MPLS-TP standards will determine how quickly PTN moves from its introductory stage into large-scale application.

References
[1] IETF RFC 5654. Requirements of an MPLS transport profile [S]. 2009.
[2] IETF RFC 3270. Multiprotocol Label Switching (MPLS) support of differentiated services [S]. 2002.
[3] IETF RFC 3985. Pseudo Wire Emulation Edge-to-Edge(PWE3) architecture [S]. 2005.
[4] IETF RFC 5254. Requirements for multi-segment Pseudo Wire Emulation Edge-to-Edge (PWE3) [S]. 2008.
[5] IETF RFC 5659. An architecture for multi-segment pseudo wire emulation edge-to-edge [S]. 2009.
[6] ITU-T Recommendation G.805(11/95). Generic functional architecture of transport networks [S]. 1995.
[7] BOCCI M, BRYANT S, LEVRAU L. A framework for MPLS in transport networks [R]. draft-ietf-mpls-tp-framework-07 (work in progress). 2009.
[8] BUSI I, ALLAN D. MPLS-TP OAM framework [R]. draft-ietf-mpls-tp-oam-framework-04 (work in progress). 2009.
[9] VIGOUREUX M, WARD D, BETTS M. Requirements for OAM in MPLS transport networks [R]. draft-ietf-mpls-tp-oam-requirements-04(work in progress). 2009.
[10] IETF RFC 2698. A two rate three color marker [S]. 1999.

[Abstract] Several key issues affect the development and standardization of Packet Transport Networks (PTN) and Multiprotocol Label Switching Transport Profile (MPLS-TP). These include the end-to-end Quality of Service (QoS) mechanism, layered network architecture, introduction of layer-3 functions, and data-plane loopback functions. This paper introduces several views on the construction and maintenance of PTN, and on the requirements of PTN services. After an analysis of Traffic Engineering (TE) based on Multiprotocol Label Switching (MPLS) and Differentiated Service (DiffServ), a service-oriented end-to-end QoS guarantee mechanism is proposed. An alternative to introducing layer-3 functions into PTN is also proposed, based on PTN layered architecture defined in the available MPLS-TP standards and drafts. Requirements of data-plane loopback functions are discussed in conclusion.