3 Replies Latest reply: Sep 22, 2017 9:24 PM by Gus RSS

    Multicast VPN: Multicast Distribution Trees

    Juan

      Multicast traffic is growing in volume in recent years and will likely continue to grow in the future. SPs and ISPs need a scalable solution to transport multicast traffic that allows them to grow in transported volume (IPTV) and in an extended portfolio of new applications that can leverage the benefits of multicast content distribution such as content delivery or telepresence in real time.

       

      On the other hand, customers need a fully integrated multicast transport service with existing unicast VPN services. The number of customers requiring multicast service between VPN sites is growing over time because some applications are ready to meet business needs efficiently using multicast transport technology.

       

      RFC 2547 and RFC 4364 only cover the MPLS L3 VPN unicast use case. The goal of Next-Generation Multicast VPN (NG mVPN) is to offer an MPLS L3 VPN integrated service, for unicast and multicast, with the next main characteristics:

       

      • Common encapsulation technology.
      • Shared control and forwarding planes.
      • Advanced Traffic Engineering (TE) capabilities.
      • Fast Restoration (FR) and protection mechanisms.
      • No need for include additional protocols in the Core (PIM avoidance).
      • Keep the Core devices free from customer related multicast state.

       

      In this discussion, we will discuss the basic concepts of Multicast Distribution Trees (MDT) in the Core and related protocols to build them, architectural differences between Classic and NG mVPN models with the most relevant differences that should be taken into account when designing mVPN solutions.



       

      Multicast Distribution Trees

       

      In this scope, Multicast Distribution Trees (MDT) are distribution trees interconnecting service provider PE devices in the core of the network.

       

      MDTs transport encapsulated customer data over the core from ingress to egress PE devices that have some mVPN in common. MDTs can be classified as inclusive and selective:

       

      • Inclusive tree: 1 MDT per mVPN. All the traffic from a given source is mapped to the same tree to be delivered from the ingress PE to all other PEs.
      • Selective tree: 1 MDT per mVPN per Source. One MDT defined for each source present in a mVPN.

       

      Selective trees are more efficient than inclusive trees because it can avoid delivering unsolicited traffic to PEs without interested receivers. In the case where it is possible to perform aggregation in the underlay, aggregate inclusive and aggregate selective trees can be useful:

       

      • Aggregate inclusive tree: 1 MDT is shared by all groups from more than one mVPN.
      • Aggregate selective tree: 1 MDT can be shared with multicast groups from more than a single mVPN.

       

       

      mcast-mvpn-mdt.png

       

       

      Figure shows the 3 main types of Core MDT along with their corresponding normative types:

       

      • Default MDT (Multidirectional Inclusive): used by a mVPN to transport signaling and low-bandwidth data flows destined to a widely distributed set of receivers
      • Data MDT (Selective): to avoid unnecessary traffic replication, the Data MDT tunnels multicast data from a Source to a defined set of interested Receivers
      • Partitioned MDT (Multidirectional Selective): a default MDT but defined with a subset of PE members of the same mVPN.

       

       

      .

      Default MDT

      Data MDT

      Partitioned MDT

      PE Connectivity

      All

      Subset

      Subset of PE

      Routing

      Bidirectional

      Unidirectional

      Both

      Availability

      Always ON

      On demand

      On demand

      Use

      Control and Data

      Data

      Data

      RFC name

      MI-PMSI

      S-PMSI

      MS-PMSI

      Signaling protocols

      PIM, Full-mesh P2MP mLDP, MP2MP mLDP, Full-mesh P2MP TE

      PIM, P2MP mLDP, P2MP TE

      PIM, Partial-mesh P2MP mLDP, MP2MP mLDP, Partial-mesh P2MP TE

       

       

       

       

      Core Trees Signaling Protocols

       

      There are a wide range of protocols that can be used to signal the MDT. The selection of one or another variant is usually driven by business needs and constrained by available resources.

       

      The main protocols for signaling the underlay core tree are: PIM, mLDP, P2MP TE and IR. Next, the main characteristics of each protocol will be discussed.

       

      PIM may be used to set up the MDT without changes. It’s exactly the same protocol used to signal source, shared or bidirectional trees and the version to be used will depend on the optimization vs. state ratio to be archived. Generally speaking, two versions could be used to set up the shared tree, PIM-SSM to set up optimal core trees at the cost of more state and PIM-BiDir to set up low state trees at the cost of suboptimal paths. With PIM in use, data replication is done at the core routers.

       

      mLDP is an enhancement to the LDP protocol (via extensions) that may be used to setup the core trees when we need for path protection features in the core (via MPLS-TE or FRR). As in the PIM case, replication is done at the core of the network. mLDP supports the creation of two kinds of trees: P2MP and MP2MP (bidirectional tree).

       

      P2MP TE protocol uses extensions to RSVP TE and the IGP to build P2MP trees in the data plane. As in the previous protocols, replication is done at the core routers. This protocol has some advantages over the previous ones if explicit routing or bandwidth reservation is a requirement. Like mLDP, P2MP TE allows for path protection.

       

      Ingress Replication (IR) is another option based on reuse the existing LSPs in the core of the network. It’s a suitable option when there are compatibility or availability limitation for deploying another option like P2MP TE or mLDP. For very simplistic scenarios, low traffic environments or simple inter-AS solutions, this may be an option to be considered. With IR the replication is done at the ingress PE and this fact is an important scalability limitation.

       

       

      Signaling protocol

      PIM

      mLDP

      P2MP TE

      IR

      State

      Soft

      Hard

      Soft

      Hard

      Encapsulation

      GRE (Rosen)

      MPLS

      MPLS

      MPLS

      Replication

      Core

      Core

      Core

      Ingress

      Core state

      Medium

      Medium

      High

      Low

      Complexity

      High

      Medium

      High

      Low

      Protection / Fast Restoration

      No

      Yes

      Yes

      Inherited

      Bandwidth reservation

      No

      No

      Yes

      Inherited

      Explicit routing

      No

      No

      Yes

      Inherited

      Best suited for

      All models

      Many-to-many

      Few-to-many

      One-to-many

      Tunneling model

      P2MP

      P2MP, MP2MP

      P2MP

      P2P

       

      I hope you find it useful.