The IP routing architecture used today in the Internet is hierarchical and represents a collection of routing domains. Below we define those routing domains.
X Routing within individual domains is provided by intra-domain routing protocols (e.g. OSPF, RIP, EIGRP).
X Routing across multiple domains is provided by inter-domain routing protocols (e.g. BGP). Routing across these inter-domain systems is usually referred to as routing between Autonomous Systems. 2
One advantage to partitioning routing into intra- and inter-domain components is the reduction in the volume of routing information that has to be maintained by routers, which is essential to providing a scalable routing system. 6 The initial partitioning of routing information is not complete. The following information describes some routing issues and resolutions:
X Transit Routing Domain – a domain that carries traffic that neither originates in the same domain nor is destined for a node in the same domain. 6
X Every router within a transit routing domain has to store in its forwarding tables all the routes provided by the inter-domain routing, regardless of whether this router is an interior router or a border router. 5
X The amount of routing information is not insignificant. This places additional demand on the resources required by the routers in order for them to maintain all necessary routes. This increase in the volume of routing information would tend to increase routing convergence time, which leads to degradation of the overall performance of the routing system. 3
X Since interior routers in a transit domain basically transfer packets, from one border router to another, it seems wasteful to have them maintain complete routing tables for all inter- and intra-domain routes.
Tag Switching provides a means by which interior routers can store only the routing information they really need. The border routers still maintain full routing information. Tag switching allows the decoupling of intra-domain and inter-domain routing, so that only TSRs at the border of a domain would be required to maintain routing information provided by the inter-domain routers. However, all other intra-domain routers in that domain would only maintain routing information associated with the interior routing protocols. Now the routing load on non-border routers has been reduced and the convergence time has been shortened.
To support this separation of interior and exterior topologies mentioned above, Tag switching allows a packet to carry not just one but a set of tags, organized in a stack.
+ Figure 6. Simple example of Tag Switching with a hierarchy of routing knowledge. 6
A TSR could either swap the tag at the top of the stack, or pop the stack, or swap the tag and push one or more tags into the stack. Inter-domains are connected via border TSRs. When a packet is forwarded between inter-domains, the tag stack in the packet contains only one tag.
When a packet is forwarded in an intra-domain, the tag stack in the packet contains two tags. The intra-domains ingress border TSR pushes the second tag onto the stack. The tag at the top of the stack provides packet forwarding information to the appropriate egress border TSR, while the next tag in the stack provides correct packet forwarding information at the egress TSR. When the packet reaches the egress TSR, or the next to last TSR, the tag stack is popped.
+ Figure 7. Example of tag stack movement within the routing hierarchy. 1
The control component used in this scenario is similar to the one used with destination-based routing. The one difference lies with the fact that this methods tag binding information is distributed both among physically adjacent TSRs and among border TSRs within a single domain. 5
Flexible Routing using Explicit Routes and QoS
Explicit routing is another extremely useful function that is supported by tag switching. In the case of destination-based routing, the destination address is the only information that is used to forward a packet. This particular function does enable highly scalable routing, but it also limits the capability to influence the actual paths taken by packets. When a need arises to evenly distribute traffic among multiple links in order to relieve the load off of those links that are over
subscribed and evenly distribute the traffic, destination-based routing is limited. Also, ISP’ s that support different classes of service will be limited due to the fact that destination-based routing will not be able to supply a separate dedicated path for that specific service. 5
Explicit routing is defined as a method of providing routes that are explicitly chosen to be other than the normal route chosen by the routing protocols. Tag switching provides support for explicit routes by using the Resource Reservation Protocol (RSVP) and defining a new RSVP Object – the Explicit Route Object.3 This Explicit Route Object is used to specify a particular explicit route. The object is carried in the RSVP PATH message. The tag binding information for the route is carried in the Tag Object by the RSVP RESV message. See RSVP below:
When a TSR wants to send an RESV message for a new RSVP flow, the TSR allocates a tag from its pool of free tags. Next, the TSR creates an entry in its TFIB with the incoming tag set to the allocated tag, places the tag in the Tag Object, and then sends out the RESV message with this object. This newly created TFIB entry contains tag information and information about local resources (e.g. queues) that packets whose tag matches the incoming tag of the entry will use. 6
The TSR populates the outgoing tag component as it receives the RESV message from its next hop TSR. Once the RSVP flow is established, the reservation state needs to be refreshed. To accomplish this, the TSR sends RESV messages associated with the flow and includes with them the same tag that the TSR bound to flow when it first created the RSVP state for the flow. This is control-driven binding. 3
The Explicit Route Object is composed of a sequence of variable-length sub-objects, where each sub-object identifies a single hop within an explicit route. The ability to express individual hops not just in terms of individual TSRs within a network topology, but in terms of a group of TSRs, provides the routing system with a significant amount of flexibility. In essence a TSR that computes an explicit route need not have detailed information about the route, whether the TSR is in the middle of the route or on the edge of the route. 6
Multicast Routing
In a multicast routing environment, multicast routing procedures are responsible for constructing a multicast distribution tree, with receivers as leaves. This tree is constructed by multicast routing protocols (e.g. DVMRP, PIM, CBT, MOSPF) and used by the forwarding component of the network layer routing to forward multicast packets. PIM is the most common protocol used in tag switching and is used in this section to describe how Tag switching supports multicast routing.
In support of multicast forwarding, each TSR associates a tag with a multicast tree as follows: 3
X A TSR creates a multicast forwarding entry, either for a shared or a source specific tree, and the list of outgoing interfaces for the entry. The TSR also creates local tags, one per outgoing interface.
X Next, the TSR creates an entry in its TFIB and populates (outgoing tag, outgoing interface, outgoing link layer information) with this information for each outgoing interface, placing a locally generated tag in the outgoing tag field. This creates a binding between a multicast tree and the tags. The TSR then advertises over each outgoing interface associated with the entry, the binding between the tag, and the tree.
X When a TSR receives a binding between a multicast tree and a tag from another TSR, if the other TSR is the upstream neighbor (with respect to the multicast tree). The local TSR places the tag carried in the binding into the incoming tag component of the TFIB entry associated with the tree.
X TSRs that are interconnected via a multiple-access subnetwork (e.g Ethernet), the tag allocation procedure for multicast has to be coordinated among the TSRs. In all other cases the tag allocation procedure for multicasting could be the same as destination-based routing.
ATM and Tag Switching
ATM forwarding is based on label swapping and the tag-switching model is also based on label swapping, tag-switching technology can be readily applied to ATM switches by implementing the control component of tag switching.
Tag information needed for tag switching is carried in the VCI field. If there were 2 levels of tagging needed, then the VPI field would be used as well. 3
To obtain the necessary control information the TSR should be able to: 8
Participate as a peer in the use of Network Layer routing protocols.
Perform routing information aggregation by supporting destination-based unicast routing in order to forward Network Layer traffic.
Support destination-based routing on an ATM switch will require the TSR to maintain several tags associated with a route, or group of routes with the same next hop. This will assist in avoiding the interleaving of packets, which arrive from different upstream tag switches, but are sent concurrently to the same next hop.
Utilize an ATM switch as a TSR and appear as a router to an adjacent router.
As stated earlier in the destination-based routing section, of the methods that accommodate tag allocation and TFIB management, upstream tag allocation and downstream tag allocation on demand are most useful in ATM networks. 4
Implementing tag switching on an ATM switch does not impede the ability to support a traditional ATM control plane (e.g. PNNI) on the same switch. Tag switching technology and the ATM control plane, would operate separately (e.g. Ships in The Night mode) with the VPI/VCI space and the other resources partitioned so that the components do not interact.
Business Aspect
In most Internet based businesses, as well as corporate enterprise networks, the rate of growth and the need for extended use of assets to become profitable makes it unrealistic for service providers (ISPs) to start building their networks from scratch with new technology. 2
Cisco’ s Tag Switching technology is a key element in its overall strategy for providing scalable networks and service solutions. Tag Switching provides network managers with the flexibility to meet current and future network designs by supporting a variety of Layer 2 technologies and Layer 3 protocols, and can be implemented or a purely routed, or switched ATM network. Cisco has also contributed the Tag Switching specification to the IETF as the basis for the emerging Multi-protocol Label Switching (MPLS) standard. 9
Today, ISPs struggle to scale existing backbone infrastructures for the future and deliver differentiated network services to save costs and generate new revenue streams. ISPs also want to be able to charge premium rates that many customers will pay for special capabilities or levels of service. Tag Switching lets ISPs: 7
Seamlessly deliver IP-based network services over high performance ATM.
Offer differentiated network services, such as QoS, and to subsequently develop and offer a price model for services.
Scale existing network infrastructures to meet future growth requirements.
Protect existing equipment investments with a Cisco IOS software only upgrade to certain ATM switches and routers.
Tag switching also complements the emerging solutions for accounting and gathering network usage statistics, and can coexist with security controls for access and resource management.
Large, enterprise backbones immediately benefit from the increased capacity and traffic management provided by Tag switching. Enterprises can also exploit Tag switching for their networks or backbones to: 7
Provide advanced QoS features that ensure network priority for mission-critical traffic.
Seamlessly integrate voice and data networks under one high-speed infrastructure.
Extend Tag-enabled ISP network services to the corporate enterprise.
Provide a more cost-effective environment by optimally using WAN bandwidth.
Scale existing enterprise backbone infrastructures to meet future requirements.
Conclusion
Of the currently proposed multilayer switching technologies, Tag Switching provides the most robust solution along with the best match with the requirements for multilayer switching. The Tag Switching topology-driven approach, which couples its control-driven destination prefix algorithm to standard routing protocols, supports much more efficient use of labels than traffic-driven per flow designs, and avoids flow-by-flow setup procedures.
One of the key innovations brought to the table by Tag Switching is the use of hierarchy of tags (e.g. Tag stacks). This enables enhancements to routing scalability by allowing FEC’ s to form a hierarchy that reflects the hierarchy in the underlying routing system.
Tag Switching also provides direct support for advanced IP services, such as CoS, RSVP, IP, VPNs, and multicast on ATM switches. This technology also brings the benefits of explicit routing and VPNs to gigabit routers. Most importantly, Cisco’ s implementation of Tag Switching is a close fit and will conform to the Multi-protocol Label Switching (MPLS) standard, both for external interfaces and for interfaces within the core of the network, enabling multi-vendor networks.
The inherent flexibility of Tag Switching also provides an outstanding match with the evolutionary requirements of public and private IP networks. Tag switching is designed from the ground up to support both packet and cell interfaces. Flexible, durable (for now) and influencing the IETF with Tag technology for the MPLS standard.
Simply stated, I believe Tag Switching, a.k.a. future name MPLS, is a viable technology able to meet the requirements brought up in the beginning of this document. Tag Switching brings layer 2 and Layer 3 functionality together with increased performance is, for right now, the best in class.
References
[1] Christopher Y. Metz , “IP Switching : Protocols and Architectures,” McGraw Hill 1999,
Chapter 11, PP. 279-307
[2] Y. Rekhter, B. Davie, D. Katz, E. Rosen, G. Swallow, et al., “Cisco Systems’ Tag Switching
Architecture Overview,” RFC 2105, February 1997.
[3] Cisco Tag Switching: Data Sheets; for further information see
http://www.cisco.com/warp/public/cc/cisco/mkt/wan/mgx8800/prodlit/tagsi_ds.htm
[4] Cisco White Paper; Tag Switching: Uniting Routing and Switching for Scalable, High-Performance
Services; see http//www.cisco.com/warp/public/cc/cisco/mkt/ios/tag/tech/tagsw_wp.htm
[5] A. Detriech,…et al., “MPLS and Cisco Tag Switching,” Proc. Cisco Networkers Conference, July
2000
[6] Bruce Davie, Yakov Rekhter, “MPLS: Technology and Applications,” Morgan Kaufman 2000,
Chapter 4, PP. 87-120
[7] Cisco IOS Software; Tag Switching:, Tag Switching: Evolving Your Network,
see http//www.cisco.com/warp/public/732/Tech/tag/overview.html
[8] Cisco On-line Documentation Paper; Tag Switching: Uniting Routing and Switching for Scalable,
High-Performance Services;
see – http//www.cisco.com/univercd/cc/td/doc/cisintwk/ito_doc/tagswtch.htm
[9] S. Agrawal, et al, “IP Switching,” Proc. Paper. for further information see
http://www.cisco.com/warp/public/cc/cisco/mkt/wan/mgx8800/prodlit/tagsi_ds.htm