Network Working Group
Request for Comments: 2340
Category: Informational

B. Jamoussi
D. Jamieson
D. Williston
S. Gabe
Nortel (Northern Telecom) Ltd.
May 1998

Nortel's Virtual Network Switching (VNS) Overview

Status of this Memo

This memo provides information for the Internet community. It does not specify an Internet standard of any kind. Distribution of this memo is unlimited.

Copyright Notice

Abstract

This document provides an overview of Virtual Network Switching (VNS).

VNS is a multi-protocol switching architecture that provides COS- sensitive packet switching, reduces the complexity of operating protocols like PPP and frame relay, provides logical networks and traffic segregation for Virtual Private Networks (VPNs), security and traffic engineering, enables efficient WAN broadcasting and multicasting, and reduces address space requirements. VNS reduces the number of routing hops over the WAN by switching packets based on labels.

VNS has been proven in production networks for several years.

   1       Introduction ............................................   2
   2       What is VNS? ............................................   3
   3       VNS Header  .............................................   5
   4       VNS Label Distribution ..................................   7
   5     Logical Networks (LNs) ....................................   7
   6       VNS Routing .............................................   8
   7       VNS Forwarding ..........................................   9
      7.1   Unicast ................................................   9
      7.2   Multicast ..............................................   9
   8       Traffic Engineering .....................................  10
   
      8.1   Equal Cost Multipaths ..................................  10
      8.2   Trunk Load Spreading ...................................  10
   9       Class of Service ........................................  11
   10      VNS Migration Strategies ................................  11
   11      Summary .................................................  11
   12      Security Considerations .................................  12
   13      Acknowledgments .........................................  12
   14      Authors' Addresses ......................................  13
   15      Full Copyright Statement ................................  14

1. Introduction

There are several key problem areas with today's wide area backbone networks that carry LAN traffic: scalability, service differentiation, redundancy, administration, and traffic containment.

First, scalability is becoming a major concern because of the rapid growth in bandwidth demand and geographical reach. As the size of the WAN network grows traditional point-to-point and NBMA topologies or network models lose their performance.

Second, the need to provide several Classes of Service (CoS) has never been greater. The days of a single "best effort" service are over and service providers demand ways to differentiate the quality of the service offered to their clients based on several policies.

Third, the WAN is often carrying mission-critical traffic and loss of service is not acceptable. So far, path redundancy has been addressed inefficiently by requiring additional links or VCs.

   Fourth, network operators demand easy and simplified network
   administration. Large NBMA topologies require extensive PVC
   provisioning until SVC  deployment becomes more ubiquitous. For
   Point-to-point models, IP address space may be used inefficiently and
   non-trivial network schemas are required to contain reserved address
   space.

Finally, proper segregation of traffic is becoming a must. This requirement is being addressed today by adding leased lines or VCs used to separate traffic flows based on regions or interest or protocol.

Nortel's Virtual Network Switching (VNS) is a technology that provides efficient solutions to these challenges.

Section 2 provides an overview of VNS. The VNS header is specified in Section 3. Section 4 describes the VNS label distribution mechanism. Section 5 defines how a VNS network can be partitioned into Logical Networks (LN). Section 6 outlines VNS routing. Section 7 defines both unicast and multicast forwarding. Section 8 describes the mechanisms used to engineer the traffic. Section 9 defines the COS based switching of VNS. Section 10 provides network migration scenarios using VNS. A summary of VNS is provided in Section 11.

2. What is VNS?

Virtual Network Switching (VNS) is a CoS-sensitive multi-protocol label switching architecture that reduces or eliminates the number of layer 3 hops over the WAN by switching traffic based on labels.

   VNS makes a network of point to point links  appear to be a single
   LAN (broadcast, multiple access) media.  The network used by a
   particular instance of VNS is called a Logical Network (LN) which is
   described in more detail in Section 5.

In reference to the ISO Network Layering Model, the Data Link Layer is expanded to include VNS network layer. To the ISO Network Layer, (e.g., IP), VNS is treated as a Data Link Layer.

           ------------------------
           | Application          |
           ------------------------
           | Presentation         |
           ------------------------
           | Session              |
           ------------------------
           | Transport            |
           ------------------------      -------------------------
           | Network (e.g., IP)   |     / Network VNS            |
           -----------------------------                         |
           | Data Link                 |--------------------------
           -----------------------------                         |
           | Physical             |     \ data link (e.g., ATM)  |
           ------------------------      -------------------------

Figure 1. ISO Network Layering Model for VNS

In a VNS Network, three separate nodal functions are defined. An ingress node, an egress node, and a tandem node. The ingress and egress nodes define the boundary between an IP network and the VNS network. Therefore, these nodes run both IP routing and VNS routing. However, tandem nodes need only run VNS routing.

   A LAN packet is encapsulated in a VNS header as it enters the LN. The
   label in the header is used to switch the packet across the LN. The
   encapsulation header contains the identifier of the last node (or
   egress node) that processes the packet as it traverses the LN. It is
   the first  node (or ingress node) that decides to which egress node
   the packet is sent. All nodes between the ingress and egress nodes
   (known as tandem nodes) decide independently the best packet
   forwarding route to the egress node identified in the packet.

The network layer protocols view VNS as a shared broadcast media, where the speed to reach any node on the media is the same for all nodes. VNS ensures that traffic destined to other nodes is forwarded optimally. This transparent view of the VNS means that all the details of the network (for example, topology and link states) can be hidden from the Upper Layer Protocols (e.g. Layer 3 routing protocols) and their applications. VNS also ensures that changes to topology and link state are hidden.

The network layer protocol on the ingress node views the network layer protocol on the egress node as its logical and directly connected neighbor. This is significant because the network layer protocols always decide which directly connected neighbor should receive a forwarded packet. The details of the actual topology supporting the connectionless network are managed entirely by the Virtual Network Switching and are hidden from the network layer protocols. To the network layer, VNS simply appears to be another Data Link Layer (or media), even though VNS is a network layer itself running on top of the actual Data Link Layer (for example, ATM trunks).

For the ingress node to choose the egress node that provides the best path to the packet's final destination, it must have knowledge of the following:

      - the nodes that can be reached in the  network
      - the topology of the network that is using the VNS services for
        transport across the network (but not necessarily the topology
        of the full network)

This knowledge is obtained through the network layer routing mechanisms such as, IP's Open Shortest Path First (OSPF) and Address Resolution Protocol (ARP).

Once the network layer protocol on the ingress node has decided which neighbor to transmit the packet to, it is the responsibility of VNS forwarding, a part of VNS, to deliver the packet to that node. Once the packet arrives at the egress node, the packet is delivered to the network layer protocol, which then forwards it to its ultimate destination.

   Tandem nodes have no interaction with the network layer protocols.
   They only require knowledge of the  VNS network topology. They make
   their packet forwarding decision on the egress node  identifier and
   LN identifier carried in the VNS header of the packet.

3. VNS Header

VNS defines a unicast header shown in Figure 2 and a multicast header shown in Figure 3.

       3                   2                   1                   0
     1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |      TTL      |      LNN            |x|LS-Key |x|DP | CmnHdr  |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    | Protocol Type |         Destination Node Identifier           |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |  COS  |x x x x|         Source Node Identifier                |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                 Network Layer Header (e.g. IP)                |
    /                                                               /
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                          Data                                 |
    /                                                               /
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Figure 2. Unicast VNS Header

The unicast header includes the following fields:

- Common Header (CmnHdr): The common header identifies the packet to be a VNS encapsulated packet.

- Discard Priority: Indicates the level of congestion at which the packet should be discarded. The value of this field is assigned on the originating node based on policy information (see Section 9).

- Load Spreading Key: indicates the stream to which the packet belongs for the purposes of equal cost multipath and trunk load spreading (see Section 8).

- LNN: The Logical Network Number defines the logical network the packet belongs to. This field in is used in conjunction with the destination node identifier as the VNS switching label (see Section 5).

- TTL: The Time To Live field is used to detect and discard packets caught in temporary routing loops.

- Destination Node Identifier: This field contains an ID which uniquely identifies the destination node. This ID is unique to the physical network not just the LN. In conjunction with the LNN, this forms a global VNS switching label.

- Protocol Type: indicates the type of Network layer protocol being carried in the packet. Examples include IP, IPX, and Bridging. If the packet is a multicast packet then this is indicated in this field.

- Source Node Identifier: This field contains an ID which uniquely identifies the source node (ingress node).

- CoS: The Class of Service field is used to provide routing class of service. The COS field also affects the Emission Priority of the packet in the scheduler (see Section 9).

Reserved Fields: All the fields marked with "x" are Reserved.

       3                   2                   1                   0
     1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |      TTL      |      LNN            |x|LS-Key |x|DP | CmnHdr  |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    | PT = Multicast|         Destination Node Identifier           |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |  COS  |x x x x|         Source Node Identifier                |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    | Protocol Type |x x x x x x x x|    Multicast Group            |
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    |                 Network Layer Header (e.g. IP)                |
    /                                                               /
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
    /                          Data                                 /
    +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

Figure 3. Multicast VNS Header

The multicast header shown in Figure 3, includes all the fields of the unicast header. In addition, the multicast header includes the following fields:

- Multicast Group: this field is used to identify a sub-group within the logical network that receives the multicast packets.

- Protocol Type: indicates the type of Network layer protocol being carried in the packet. Examples include IP, IPX, and Bridging.

4. VNS Label Distribution

Label distribution in VNS is based on a distributed serverless topology driven approach. Standard ARP or address gleaning is used to distribute and map network layer addresses to VNS addresses.

A VNS Label is an 6 byte encoding of the LNN and the node ID. VNS Labels are treated as MAC addresses by the network layer. This means that labels are distributed by the same means network layers use to distribute MAC addresses. Thus, VNS leverages existing L2/L3 mapping techniques and doesn't require a separate Label Distribution Protocol.

5. Logical Networks (LNs)

A logical network consists of a subset of the nodes in a network together with a subset of the trunking facilities that link those nodes. Logical networks partition the network into subnetworks that serve a subset of the overall topology.

Each of the logical networks supported on any given node has a separate routing and forwarding table (built by VNS). Therefore, routing decisions are based on the resources available to the logical network, not the entire network.

Each instance of VNS will discover all the trunks which are connected to neighbors which support a matching LNN. This provides a huge administrative saving, since VNS provisioning is on a per-node basis, not on a per-link basis. VNS provisioning requires only a unique node ID and an LNN. Discovery of which trunks support which LNNs is done at run time, relieving administrative effort, and allowing the LN to dynamically adapt to topology changes.

Multiple Logical Networks provide the following benefits to the network system:

- Logical networks allow service providers to service multiple private networks or (Virtual Private Internets) easily over one network.

- Logical networks can be used to limit the impact of one network layer protocol on the others. This is particularly true for protocols that broadcast or multicast a large percentage of either their control or data packets. This increases the effective bandwidth of the trunks and allows the overall network to scale better.

- Logical networks allow for the configuration of the network to meet individual community of interest and geographical subnetworking needs.

- Routing control traffic has significance only in the local subnetwork that is isolated to that subnetwork.

- Logical networks allow different instances of the same protocol to share trunk facilities.

6. VNS Routing

VNS routing is a link state routing system which uses many concepts similar to OSPF and PNNI. One of the most significant departures from the others is its ability to calculate shortest path trees for routing unicast traffic and spanning trees for routing multicast traffic within a Logical Network.

There is only one type of interface that VNS routing supports and this is known as a VNS link. A link is a set of trunks that join two VNS neighbor nodes. Each node in a VNS network maintains information about the state of locally attached links. This information is flooded throughout the network whenever there is a significant change to the link's state or attributes (i.e. up/down, speed change, available bandwidth change).

Each node stores and forwards the link state information received from all other nodes. This allows each node to have the same view of all of the nodes in the network together with all of their link state information. This data is used to compute both the shortest path to reach each node in the Logical Network and a spanning tree for the Logical Network.

Logical networks are not bound to a particular trunk or link. They are configured on a node. By default, a link will support a specific logical network if the two nodes which it connects both are configured to support the logical network number. This provides a significant savings in operations over having to configure logical networks on links or trunks.

When a link first comes into service, a protocol is run which allows the two neighboring nodes to exchange information about the logical networks they support. This allows the two nodes to determine if the links are to be considered as a locally attached link for a logical network.

7. VNS Forwarding

VNS supports two types of forwarding: unicasting and multicasting. In the first type, the data packet arrives on the ingress node and unicasting forwards the data packet to a single destination (egress node). In the second type, the data packet arrives on the ingress node and multicasting forwards the data packet to all other nodes in the logical network.

7.1 Unicast

   When a packet first enters the  LAN internetwork, the network layer
   routing protocol determines the next hop of the best route for the
   packet to reach its final destination. If the best route is through a
   VNS Logical Network, the network layer routing protocol relies on VNS
   forwarding to get the packet to the egress  node. A VNS packet header
   containing the node ID (the unique ID assigned  to each  node) of the
   egress node is added to the front of the packet and VNS forwarding is
   invoked to deliver the packet. The network layer routing protocol
   learns the egress node ID through an Address Resolution Protocol
   (ARP) for IP and Source Address learning for bridging.

As the packet traverses the LN, routing decisions are made to determine the next hop in the route to reach the destination node ID specified in the VNS header. A forwarding table is built on each node that assists in making the routing decision.

   Each VNS instance on each  node builds and maintains a forwarding
   table for its LN. Each forwarding table has an entry for every  node
   that is a member of the logical network.

7.2 Multicast

In addition to the unicast forwarding function, VNS also supports a multicast forwarding service for traffic within an LN at the VNS layer. Multicast packets are delivered to all nodes supporting the logical network to which the multicast packet belongs. The packets are sent along the branches of a spanning tree that is built by each node supporting the logical network and is based on a common root node (so that each node's view of the tree is the same as other nodes). In other words, multicast packets are sent intelligently, consuming a minimum of network bandwidth. If the network topology is stable, each node receives each multicast packet only once.

Multicast packets received at any node are not acknowledged. They are simply forwarded to the specified network layer interface and sent to any other neighbor nodes on the spanning tree.

8. Traffic Engineering

VNS forwarding supports two types of traffic engineering mechanisms: equal cost multipaths and trunk load spreading.

Equal cost multipaths allows different streams (unique network layer source and destination address pairings) to be load spread between multiple relatively equal cost paths, through the Logical Network to the egress node.

   Trunk load spreading between two neighbors can take place when
   multiple VNS  trunks are defined between neighbors. Again, the load
   spreading is based on network layer streams.

8.1 Equal Cost Multipaths

From any point in a logical network, there may be multiple paths to reach a specific egress node. If VNS routing determines that more than one of these paths are of equal cost, VNS packets will be load spread between two of them.

Equal cost multipath forwarding is supported not only on ingress nodes but on tandem nodes as well. Each packet on an ingress node is tagged with an equal cost multipath key. This key is acted upon at the ingress node and stored in the VNS header to be used on tandem nodes.

The equal cost multipath key is calculated by running an algorithm over the source and destination network layer addresses. This means that, in a stable network, any given stream will always take the same path through a Logical Network avoiding the problems that misordering would otherwise cause.

8.2 Trunk Load Spreading Between Neighbors

VNS allows multiple trunks to be configured between neighboring VNS nodes. VNS routing considers the aggregate bandwidth of those trunks to determine the metric between the nodes. Also, VNS load spreads its traffic amongst those trunks.

As is the case with equal cost multipaths, the trunk load spreading key is calculated on the ingress node from an algorithm run over the source and destination network layer addresses. The key is then stored in the VNS header to be used on all tandem nodes through the Logical Network.

9. Class of Service

   At the ingress to a VNS Network, packets are classified according to
   the Class of Service (Cos) policy settings. The CoS differentiation
   is achieved through different  Emission and Discard priorities. The
   semantics of the classification is carried in the VNS label (DP and
   COS Fields described in Section 3) to be used at the ingress node as
   well as all tandem points in the VNS network to affect queuing and
   scheduling decisions.

10. VNS Migration Strategies

   VNS supports several upper layer protocols such as IP, IPX, and
   Bridging. Therefore, it is a multiprotocol label switching
   architecture. In addition, VNS  is not tied to a particular L2
   technology. It runs on cell (e.g., ATM) trunks, frame trunks, or a
   mixture of both.

VNS can be gradually introduced in a network. It can be implemented between switching elements interconnected by point to point links. Each of the switching nodes can run layer 3 routing simultaneously with packet switching. VNS also allows for the interconnection of VNS clouds through an ATM VC.

Since VNS can run on a mixture of Frame and Cell trunks, it allows for the graceful migration of the frame links to ATM without requiring a complete immediate overhaul.

11. Summary

VNS addresses scalability problems in several ways:

      1. By a generally distributed design which doesn't
         require a Label Distribution Protocol, or servers of any kind.
      2. By providing an efficient, distributed multicast mechanism.
      3. By allowing administrators to control the size of a
         Logical Network, limiting traffic to a subset of the physical
         topology.
      4. By reducing layer 3 address space/subnet requirements in the
         WAN which reduces the routing table size.

VNS provides redundancy transparent to the network layer protocol by managing the network of trunks independently of the network layer. VNS will automatically discover any topology changes and re-route traffic accordingly.

VNS eases network administration by dynamically keeping track of which trunks are available for each LNN. Network administrators don't have to configure VNS or network layer addresses on a per link basis. Network layer addresses only have to be assigned on a per Logical Network basis. For nodes which will only be tandem VNS nodes, network layer addresses aren't required at all.

Since VNS traffic is constrained within an LNN, administrators have control of where VNS traffic is allowed to flow.

   Finally, VNS supports switching of several Upper Layer Protocols and
   supports  several media (cell and Frame) or a mixture thereof.
   Switching in the core of the WAN removes the need for routers and
   improves the performance due to a reduction in the  number of fields
   that need to processed.

12. Security Considerations

Logical networks provide a means of restricting traffic flow for security purposes. VNS also relies on the inherent security of the L2 media such as an ATM Virtual Circuit.

13. Acknowledgments

The authors would like to acknowledge the valuable comments of Terry Boland, Pierre Cousineau, Robert Eros, Robert Tomkins, and John Whatman.

14. Authors' Addresses

Bilel Jamoussi
Nortel (Northern Telecom), Ltd.
PO Box 3511 Station C
Ottawa ON K1Y 4H7
Canada

EMail:

          jamoussi@Nortel.ca

Dwight Jamieson
Nortel (Northern Telecom), Ltd.
PO Box 3511 Station C
Ottawa ON K1Y 4H7
Canada

EMail:

          djamies@Nortel.ca

Dan Williston
Nortel (Northern Telecom), Ltd.
PO Box 3511 Station C
Ottawa ON K1Y 4H7
Canada

EMail:

          danwil@Nortel.ca

Stephen Gabe
Nortel (Northern Telecom), Ltd.
PO Box 3511 Station C
Ottawa ON K1Y 4H7
Canada

EMail:

          spgabe@Nortel.ca

15. Full Copyright Statement

This document and translations of it may be copied and furnished to others, and derivative works that comment on or otherwise explain it or assist in its implementation may be prepared, copied, published and distributed, in whole or in part, without restriction of any kind, provided that the above copyright notice and this paragraph are included on all such copies and derivative works. However, this document itself may not be modified in any way, such as by removing the copyright notice or references to the Internet Society or other Internet organizations, except as needed for the purpose of developing Internet standards in which case the procedures for copyrights defined in the Internet Standards process must be followed, or as required to translate it into languages other than English.

The limited permissions granted above are perpetual and will not be revoked by the Internet Society or its successors or assigns.

This document and the information contained herein is provided on an "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.