Evaluation of Existing GMPLS Encoding against G.709v3 Optical Transport Networks (OTNs)

GMPLS routing and signaling provide the mechanisms for basic GMPLS control of Optical Transport Networks (OTNs) based on the 2001 revision of the G.709 specification . The 2012 revision of the G.709 specification includes new OTN features that are not supported by GMPLS. This document provides an evaluation of exiting GMPLS signaling and routing protocols against G.709 requirements. Background information and a framework for the GMPLS protocol extensions needed to support G.709 is provided in . Specific routing and signaling extensions defined in and specifically address the gaps identified in this document.

The digital OTN-layered structure is comprised of the digital path layer (ODU) and the digital section layer (OTU). An OTU (Optical channel Transport Unit) section layer supports one ODU path layer as a client and provides monitoring capability for the Optical Channel (OCh), which is the optical path carrying the digital OTN structure. An ODU path layer may transport a heterogeneous assembly of ODU clients. Some types of ODUs (i.e., ODU1, ODU2, ODU3, and ODU4) may assume either a client or server role within the context of a particular networking domain. The terms ODU1, ODU2, ODU3, ODU4, and flexible ODU (ODUflex) are explained in G.709. G.872 provides two tables defining mapping and multiplexing capabilities of OTNs, which are reported below.

In the following, the terms Optical channel Data Unit-j (ODUj) and Optical channel Data Unit-k (ODUk) are used in a multiplexing scenario to identify the lower order signal (ODUj) and the higher order signal (ODUk). How an ODUk connection service is transported within an operator network is governed by operator policy. For example, the ODUk connection service might be transported over an ODUk path over an Optical channel Transport Unit-k (OTUk) section, with the same path and section rates as that of the connection service (see ). In this case, an entire lambda of capacity is consumed in transporting the ODUk connection service. On the other hand, the operator might exploit different multiplexing capabilities in the network to improve infrastructure efficiencies within any given networking domain. In this case, ODUk multiplexing may be performed prior to transport over various rate ODU servers (as per ) over associated OTU sections. From the perspective of multiplexing relationships, a given ODUk may play different roles as it traverses various networking domains. As detailed in , client ODUk connection services can be transported over: one or more wavelength subnetworks connected by optical links, or one or more ODU links (having sub-lambda and/or lambda bandwidth granularity), or a mix of ODU links and wavelength subnetworks. This document considers the Traffic Engineering (TE) information needed for ODU path computation and the parameters needed to be signaled for Label Switched Path (LSP) setup. The following sections list and analyze what GMPLS already has and what it is missing with regard to each type of data that needs to be advertised and signaled.

G.709 defines two types of Tributary Slot (TS) granularities. This TS granularity is defined per layer, meaning that both ends of a link can select proper TS granularity differently for each supported layer, based on the rules below: If both ends of a link are new cards supporting both 1.25 Gbit/s TS and 2.5 Gbit/s TS, then the link will work with 1.25 Gbit/s TS. If one end of a link is a new card supporting both the 1.25 Gbit/s and 2.5 Gbit/s TS granularities, and the other end is an old card supporting just the 2.5 Gbit/s TS granularity, the link will work with 2.5 Gbit/s TS granularity.

As defined in G.709, an ODUk container consists of an Optical channel Payload Unit-k (OPUk) plus a specific ODUk Overhead (OH). OPUk OH information is added to the OPUk information payload to create an OPUk. It includes information to support the adaptation of client signals. Within the OPUk overhead, there is the payload structure identifier (PSI) that includes the payload type (PT). The PT is used to indicate the composition of the OPUk signal. When an ODUj signal is multiplexed into an ODUk, the ODUj signal is first extended with the frame alignment overhead and then mapped into an Optical channel Data Tributary Unit (ODTU). Two different types of ODTUs are defined: ODTUjk ((j,k) = {(0,1), (1,2), (1,3), (2,3)}; ODTU01, ODTU12, ODTU13, and ODTU23) in which an ODUj signal is mapped via the Asynchronous Mapping Procedure (AMP), as defined in Section 19.5 of . ODTUk.ts ((k,ts) = (2,1..8), (3,1..32), (4,1..80)) in which a lower order ODU (ODU0, ODU1, ODU2, ODU2e, ODU3, and ODUflex) signal is mapped via the Generic Mapping Procedure (GMP), as defined in Section 19.6 of . G.709 also introduces a logical entity, called Optical channel Data Tributary Unit Group (ODTUGk), characterizing the multiplexing of the various ODTU. The ODTUGk is then mapped into OPUk. Optical channel Data Tributary Unit j into k (ODTUjk) and Optical channel Data Tributary Unit k with ts tributary slots (ODTUk.ts) are directly time-division multiplexed into the tributary slots of an OH OPUk. When PT is assuming values 0x20 or 0x21, together with OPUk type (k=1, 2, 3, 4), it is used to discriminate two different ODU multiplex structures for ODTUGx: Value 0x20: supporting ODTUjk only Value 0x21: supporting ODTUk.ts or ODTUk.ts and ODTUjk The distinction is needed for OPUk with k=2 or 3 since OPU2 and OPU3 are able to support both the different ODU multiplex structures. For OPU4 and OPU1, only one type of ODTUG is supported: ODTUG4 with PT=0x21 and ODTUG1 with PT=0x20 (see ). The relationship between PT and TS granularity is due to the fact that the two different ODTUGk types discriminated by PT and OPUk are characterized by two different TS granularities of the related OPUk, the former at 2.5 Gbit/s and the latter at 1.25 Gbit/s. In order to complete the picture, in the PSI OH, there is also the Multiplex Structure Identifier (MSI) that provides the information on which tributary slots of the different ODTUjk or ODTUk.ts are mapped into the related OPUk. The following figure shows how the client traffic is multiplexed till the OPUk layer.

G.798 describes the so-called PT=0x21-to-PT=0x20 interworking process that explains how two nodes with interfaces that have different payload types and, hence, different TS granularity (1.25 Gbit/s vs. 2.5 Gbit/s), can be coordinated to permit the equipment with 1.25 Gbit/s TS granularity to adapt the TS allocation according to the different TS granularity (2.5 Gbit/s) of a neighbor. Therefore, in order to let the Network Element (NE) change TS granularity accordingly to the neighbor requirements, the AUTOpayloadtype needs to be set. When both the neighbors (link or trail) have been configured as structured, the payload type received in the overhead is compared to the transmitted PT. If they are different and the transmitted one is PT=0x21, the node must fall back to PT=0x20. In this case, the fallback process makes the system self-consistent, and the only reason for signaling the TS granularity is to provide the correct label (i.e., the label for PT=0x21 has twice the TS number of PT=0x20). On the other side, if the AUTOpayloadtype is not configured, the Resource Reservation Protocol-Traffic Engineering (RSVP-TE) consequent actions need to be defined in case of a TS mismatch.

When setting up an ODUj over an ODUk, it is possible to identify two types of TS granularity (TSG): the server and the client. The server TS granularity is used to map an end-to-end ODUj onto a server ODUk LSP or links. This parameter cannot be influenced in any way from the ODUj LSP: the ODUj LSP will be mapped on tributary slots available on the different links / ODUk LSPs. When setting up an ODUj at a given rate, the fact that it is carried over a path composed by links / Forwarding Adjacencies (FAs) structured with 1.25 Gbit/s or 2.5 Gbit/s TS granularity is completely transparent to the end-to-end ODUj. The client TS granularity information is one of the parameters needed to correctly select the adaptation towards the client layers at the end nodes, and this is the only thing that the ODUj has to guarantee. In , an example of client and server TS granularity utilization in a scenario with mixed OTN and OTN interfaces is shown.

In this scenario, an ODU3 LSP is set up from nodes B to Z. Node B has an old interface that is able to support 2.5 Gbit/s TS granularity; hence, only client TS granularity equal to 2.5 Gbit/s can be exported to ODU3 H-LSP-possible clients. An ODU2 LSP is set up from nodes A to Z with client TS granularity 1.25 Gbit/s signaled and exported towards clients. The ODU2 LSP is carried by ODU3 H-LSP from nodes B to Z. Due to the limitations of the old node B interface, the ODU2 LSP is mapped with 2.5 Gbit/s TS granularity over the ODU3 H-LSP. Then, an ODU1 LSP is set up from nodes A to Z, which is carried by the ODU2 H-LSP and mapped over it using 1.25 Gbit/s TS granularity. What is shown in the example is that the TS granularity processing is a per-layer issue: even if the ODU3 H-LSP is created with the TS granularity client at 2.5 Gbit/s, the ODU2 H-LSP must guarantee a 1.25 Gbit/s TS granularity client. The ODU3 H-LSP is eligible from an ODU2 LSP perspective since it is known from the routing that this ODU3 interface at node Z supports an ODU2 termination exporting a TS granularity at 1.25 Gbit/s / 2.5 Gbit/s. The TS granularity information is needed in the routing protocol as the ingress node (A in the previous example) needs to know if the interfaces at the last hop can support the required TS granularity. In case they cannot, A will compute an alternate path from itself to Z (see ). Moreover, TS granularity information also needs to be signaled. As an example, consider the setup of an ODU3 forwarding adjacency that is going to carry an ODU0; hence, the support of 1.25 Gbit/s TS is needed. The information related to the TS granularity has to be carried in the signaling to permit node C (see ) to choose the right one among the different interfaces (with different TS granularities) towards D. In case the full Explicit Route Object (ERO) is provided in the signaling with explicit interface declaration, there is no need for C to choose the right interface towards D as it has been already decided by the ingress node or by the Path Computation Element (PCE).

ODU0 <--------------------------------------> | | +--------+ +--------+ +--------+ +--------+ | | | | | | 1.25 | | | Node | | Node | | Node +------+ Node | | A +------+ B +------+ C | ODU3 | D | | | ODU3 | | ODU3 | +------+ | +--------+ 1.25 +--------+ 2.5 +--------+ 2.5 +--------+ ]]> In case an ODUk FA_LSP needs to be set up as nesting another ODUj (as depicted in ), there might be the need to know the hierarchy of nested LSPs in addition to TS granularity to permit the penultimate hop (i.e., C) to choose the correct interface towards the egress node or any intermediate node (i.e., B) to choose the right path when performing the ERO expansion. This is not needed in case we allow bundling only component links with homogeneous hierarchies. In the case in which a specific implementation does not specify the last hop interface in the ERO, crankback can be a solution. In a multi-stage multiplexing environment, any layer can have a different TS granularity structure; for example, in a multiplexing hierarchy such as ODU0->ODU2->ODU3, the ODU3 can be structured at TS granularity = 2.5 Gbit/s in order to support an ODU2 connection, but this ODU2 connection can be a tunnel for ODU0 and, hence, structured with 1.25 Gbit/s TS granularity. Therefore, any multiplexing level has to advertise its TS granularity capabilities in order to allow a correct path computation by the end nodes (both the ODUk trail and the H-LSP/FA). The following table shows the different mapping possibilities depending on the TS granularity types. The client types are shown in the left column, while the different OPUk server and related TS granularities are listed in the top row. The table also shows the relationship between the TS granularity and the payload type.

Specific information could be defined in order to carry the multiplexing hierarchy and adaptation information (i.e., TS granularity / PT and AMP / GMP) to enable precise path selection. That way, when the penultimate node (or the intermediate node performing the ERO expansion) receives such an object, together with the Traffic Parameters Object, it is possible to choose the correct interface towards the egress node. In conclusion, both routing and signaling need to be extended to appropriately represent the TS granularity/PT information. Routing needs to represent a link's TS granularity and PT capabilities as well as the supported multiplexing hierarchy. Signaling needs to represent the TS granularity/PT and multiplexing hierarchy encoding.

supports only the deprecated auto-MSI mode, which assumes that the Tributary Port Number (TPN) is automatically assigned in the transmit direction and is not checked in the receive direction. As described in and , the OPUk overhead in an OTUk frame contains n (n = the total number of TSs of the ODUk) MSI bytes (in the form of multiframe), each of which is used to indicate the association between the TPN and TS of the ODUk. The association between the TPN and TS has to be configured by the control plane and checked by the data plane on each side of the link. (Please refer to for further details.) As a consequence, the RSVP-TE signaling needs to be extended to support the TPN assignment function.

From a routing perspective, GMPLS OSPF and GMPLS IS-IS only allow advertising interfaces (the single TS type) without the capability of providing precise information about bandwidth-specific allocation. For example, in case of link bundling, when dividing the unreserved bandwidth by the MAX LSP bandwidth, it is not possible to know the exact number of LSPs at MAX LSP bandwidth size that can be set up (see the example in ). The lack of spatial allocation heavily impacts the restoration process because the lack of information on free resources highly increases the number of crankbacks affecting network convergence time. Moreover, actual tools provided by and only allow advertising signal types with fixed bandwidth and implicit hierarchy (e.g., Synchronous Digital Hierarchy (SDH) networks / Synchronous Optical Networks (SONETs)) or variable bandwidth with no hierarchy (e.g., packet switching networks); but, they do not provide the means for advertising networks with a mixed approach (e.g., ODUflex Constant Bit Rate (CBR) and ODUflex packet). For example, when advertising ODU0 as MIN LSP bandwidth and ODU4 as MAX LSP bandwidth, it is not possible to state whether the advertised link supports ODU4 and ODUflex or ODU4, ODU3, ODU2, ODU1, ODU0, and ODUflex. Such ambiguity is not present in SDH networks where the hierarchy is implicit and flexible containers like ODUflex do not exist. The issue could be resolved by declaring 1 Interface Switching Capability Descriptor (ISCD) for each signal type actually supported by the link. Suppose, for example, there is an equivalent ODU2 unreserved bandwidth in a TE link (with bundling capability) distributed on 4 ODU1; it would be advertised via the ISCD in this way: MAX LSP Bandwidth: ODU1 MIN LSP Bandwidth: ODU1 - Maximum Reservable Bandwidth (of the bundle) set to ODU2 - Unreserved Bandwidth (of the bundle) set to ODU2 In conclusion, the routing extensions defined in and require a different ISCD per signal type in order to advertise each supported container. This motivates an attempt to look for a more optimized solution without proliferation of the number of ISCDs advertised. Per , OSPF messages are directly encapsulated in IP datagrams and depend on IP fragmentation when transmitting packets larger than the network's MTU. recommends that "IP fragmentation should be avoided whenever possible". This recommendation further constrains solutions since OSPF does not support any generic mechanism to fragment OSPF Link State Advertisements (LSAs). Even when used in IP environments, IS-IS does not support message sizes larger than a link's maximum frame size. With respect to link bundling , the utilization of the ISCD as it is would not allow precise advertising of spatial bandwidth allocation information unless using only one component link per TE link. On the other hand, from a signaling point of view, describes GMPLS signaling extensions to support the control of G.709 OTNs defined before 2011 . However, needs to be updated because it does not provide the means to signal all the new signal types and related mapping and multiplexing functionalities.

In the current traffic parameters signaling, bit rate and tolerance are implicitly defined by the signal type. ODUflex CBR and ODUflex packet can have variable bit rates (please refer to , Table 2); hence, signaling traffic parameters need to be upgraded. With respect to tolerance, there is no need to upgrade GMPLS protocols as a fixed value (+/-100 parts per million (ppm) or +/-20 ppm depending on the signal type) is defined for each signal type.

Unreserved resources need to be advertised per priority and per signal type in order to allow the correct functioning of the restoration process. only allows advertising unreserved resources per priority; this leads to uncertainty about how many LSPs of a specific signal type can be restored. As an example, consider the scenario depicted in the following figure.

Consider the case where a TE link is composed of three ODU3 component links with 32 TSs available on the first one, 24 TSs on the second, and 24 TSs on the third and is supporting ODU2 and ODU3 signal types. The node would advertise a TE link with unreserved bandwidth equal to 80 TSs and a MAX LSP bandwidth equal to 32 TSs. In case of restoration, the network could try to restore two ODU3s (64 TSs) in such a TE link while only a single ODU3 can be set up, and a crankback would be originated. In more complex network scenarios, the number of crankbacks can be much higher.

Maximum LSP bandwidth is currently advertised per priority in the common part of the ISCD. Section 5 reviews some of the implications of advertising OTN information using ISCDs and identifies the need for a more optimized solution. While strictly not required, such an optimization effort should also consider the optimization of the per-priority maximum LSP bandwidth advertisement of both fixed and variable ODU types.

The capability advertised by an interface needs further distinction in order to separate terminating and switching capabilities. Due to internal constraints and/or limitations, the type of signal being advertised by an interface could just be switched (i.e., forwarded to the switching matrix without multiplexing/demultiplexing actions), terminated (demultiplexed), or both. The following figures help explain the switching and terminating capabilities.

| ODU2 |----|----------|--------\ | | +-------+ | | +----+ | | | | \__/ | | | | \/ | | +-------+ | ODU3 | | ODU3 | ----->| ODU3 |----|----------|------\ | | | +-------+ | | \ | | | | | \| | | | | +----+ | | | | \__/ | | | | \/ | | | | ---------> OTU3 +-----------------+ +-----------------+ ]]> The figure in the example shows a line interface that is able to: Multiplex an ODU2 coming from the switching matrix into an ODU3 and map it into an OTU3 Map an ODU3 coming from the switching matrix into an OTU3 In this case, the interface bandwidth advertised is ODU2 with switching capability and ODU3 with both switching and terminating capabilities. This piece of information needs to be advertised together with the related unreserved bandwidth and signal type. As a consequence, signaling must have the capability to set up an LSP, allowing the local selection of resources to be consistent with the limitations considered during the path computation. In and , there are two examples of the terminating/switching capability differentiation. In both examples, all nodes only support single-stage capability. represents a scenario in which a failure on link B-C forces node A to calculate another ODU2 LSP carrying ODU0 service along the nodes B-E-D. As node D is a single stage capable node, it is able to extract ODU0 service only from the ODU2 interface. Node A has to know that from E to D exists an available OTU2 link from which node D can extract the ODU0 service. This information is required in order to avoid the OTU3 link being considered in the path computation.

| A |_________| B |_________| C |_________| D |----> | | | | | | | | +-----+ +--+--+ +-----+ ++--+-+ | | | OTU3| | | Link| +-----+__________________| | | | | OTU3 Link | |____| E | | | |_____________________| +-----+ OTU2 Link ]]> addresses the scenario in which the restoration of the ODU2 LSP (A-B-C-D) is required. The two bundled component links between B and E could be used, but the ODU2 over the OTU2 component link can only be terminated and not switched. This implies that it cannot be used to restore the ODU2 LSP (A-B-C-D). However, such ODU2 unreserved bandwidth must be advertised since it can be used for a different ODU2 LSP terminating on E, e.g., F-B-E. Node A has to know that the ODU2 capability on the OTU2 link can only be terminated, and that the restoration of A-B-C-D can only be performed using the ODU2 bandwidth available on the OTU3 link.

| A |_________| B |_________| C |_________| D |----> | | | | | | | | +-----+ ++-+-++ +-----+ +--+--+ | | | | OTU2| | | | +-----+ Link| | | OTU3 +-----+ | | | | | | Link | | | | F |_______| | |___________| E |___________| | | |_____________| | OTU2 Link +-----+ OTU2 Link +-----+ ]]> The issue shown above is analyzed in an OTN context, but it is a general technology-independent GMPLS limitation.

defines eight priorities for resource availability and usage. As defined, each is advertised independent of the number of priorities supported by a network, and even unsupported priorities are included. As is the case in , addressing any inefficiency with such advertisements is not required to support OTNs. But, any such inefficiency should also be considered as part of the optimization effort identified in .

With reference to , the introduction of multi-stage multiplexing implies the advertisement of cascaded adaptation capabilities together with the matrix access constraints. The structure defined by the IETF for the advertisement of adaptation capabilities is the Interface Adaptation Capability Descriptor (IACD), as defined in . With respect to routing, please note that in case of multi-stage multiplexing hierarchy (e.g., ODU1->ODU2->ODU3), not only the ODUk/OTUk bandwidth (ODU3) and service-layer bandwidth (ODU1) are needed but also the intermediate one (ODU2). This is a typical case of a spatial allocation problem. In this scenario, suppose the following advertisement: Hierarchy: ODU1->ODU2->ODU3 Number of ODU1==5 The number of ODU1 suggests that it is possible to have an ODU2 FA, but it depends on the spatial allocation of such ODU1s. It is possible that two links are bundled together and three ODU1->ODU2->ODU3 are available on a component link and two on the other one; in such a case, the ODU2 FA could not be set up. The advertisement of the ODU2 is needed because in case of ODU1 spatial allocation (3+2), the ODU2 available bandwidth would be 0 (ODU2 FA cannot be created), while in case of ODU1 spatial allocation (4+1), the ODU2 available bandwidth would be 1 (1 ODU2 FA can be created). The information stated above implies augmenting both the ISCD and the IACD.

The ODUk label format defined in could be updated to support new signal types as defined in , but it would be difficult to further enhance it to support possible new signal types. Furthermore, such a label format may have scalability issues due to the high number of labels needed when signaling large LSPs. For example, when an ODU3 is mapped into an ODU4 with 1.25 Gbit/s tributary slots, it would require the utilization of 31 labels (31*4*8=992 bits) to be allocated, while an ODUflex into an ODU4 may need up to 80 labels (80*4*8=2560 bits). A new flexible and scalable ODUk label format needs to be defined.

This document provides an evaluation of OTN requirements against actual routing (, , and ) and signaling mechanisms (, , and ) in GMPLS. This document defines new types of information to be carried that describes OTN containers and hierarchies. It does not define any new protocol elements, and from a security standpoint, this memo does not introduce further risks with respect to the information that can be currently conveyed via GMPLS protocols. For a general discussion on MPLS and GMPLS-related security issues, see the MPLS/GMPLS security framework .

The authors would like to thank Lou Berger, Eve Varma, and Sergio Lanzone for their precious collaboration and review.