IPv6 Migration Considerations
By Brian Gottbetter, Director of IP Product Management and Mark Bath, IP Product Development Engineer
Global Crossing

One of the early considerations of an IPv6 migration should be how to implement IPv6 packet forwarding and routing methodology. All edge devices in the network will need to be dual stack enabled (this means running an IPv4 and IPv6 protocol stack on the router in parallel). But ask yourself, does the core of the IP network need to be IPv6 enabled? If a tunneling mechanism, such as 6PE (the ability to forward IPv6 traffic via MPLS), or some form of IPinIP tunneling such as GRE is used, the core of the network will not need to be IPv6 aware, thus negating the need for the core network to be implemented in a dual stack configuration.

There are numerous reasons why a tunneled solution might be of an initial advantage. Deployment of anything new needs to be controlled, and its impacts on existing network services considered. The careful deployment of a few dedicated IPv6 access devices using an IPinIP tunneling mechanism through the core enables a very controlled implementation. It allows the business to gain operational experience of running and managing an IPv6 network without impacting the IPv4 service offerings of the network. Such a solution also permits additional testing, namely resolving any bugs and problems with the router code or configuration standards. Another benefit to the business of this initial deployment is that Operational Support Systems used to run and manage the network can be modified to support IPv6 networking.

Global Crossing ran such a network topology for nearly four years, allowing it to achieve all of the above, with no impact to its existing IP services portfolio based on IPv4. It also gave its customer base the ability to gain similar experience.

This concept works in a limited environment but does not really scale to the size of deployment Global Crossing, any other network provider or corporate organization is likely to require. Maintaining a full mesh of IPinIP tunnels is expensive, both operationally and in terms of router load. In order to run a seamless IPv4 and IPv6 network, the goal should be to ensure that all access routers can support dual stack, and therefore IPv6 nodes and devices that are attached. In some cases these access routers may be low-end devices with limited CPU resources to manage the IPinIP tunneling. So what are the options?

The only real options for a full scale deployment are to run dual stack through the entire network or some other tunneling protocol that is not so resource intensive on the access router such as 6PE. Unfortunately there is no right or wrong answer. The final design is really going to be driven by requirements and costs. In fact the likelihood is that some element of dual stack, 6PE and even IPinIP tunneling may all factor into the final design.

If the core network is already MPLS aware and able to forward MPLS encapsulated packets then 6PE has a major advantage over Dual stack. The deployment of IPv6 services throughout the entire network is only dependant on the edge; the core will just see MPLS encapsulated packets, which it is already able to handle. This means that there are no code upgrades and potentially associated hardware upgrades needed in the core. If the core is not yet MPLS enabled then a choice has to be made, to dual stack or to deploy MPLS and 6PE.

If MPLS is already in your network development plans, then making a dependency on IPv6 deployment requiring the completion of the MPLS deployment may be one option. If there are no MPLS plans, then a comparison exercise should be undertaken to investigate the efforts and impacts of deploying MPLS in the core or dual stack. A number of factors need to be considered that include hardware availability, operational impacts and business impacts.

For example if current hardware and software in the core network can handle MPLS with no upgrades, yet the deployment of dual stack requires software and memory upgrades, and costs is a controlling factor, then the likely decision would be to take the 6PE route in the core rather than dual stack.

Many business managers’ primary objective is to leverage assets as efficiently as possible to achieve the lowest Total Cost of Ownership. This means that a lot of devices within the core network may already be maxing out or near maximum in terms of memory and CPE utilization, There is also a good chance that the code level of the existing routers will not support IPv6, thus a software refresh would be needed. With routers potentially already running near to their maximum, new code may just tip the balance requiring additional capital expenditure. Refreshing of code always requires more resources from the router, as the new code release is likely to have lots of new features besides IPv6.

In addition, (to the new code) is the overhead of maintaining an IPv6 routing table and relevant routing protocols along side the current IPv4 setup. Hopefully the only upgrade needed is the addition of more memory or a faster processor. However, if this is not possible then a complete new router may be needed. It should be noted that these costs issues are not just associated with the core, but also the edge access devices. A complete network upgrade or refresh is always more expensive than a partial upgrade, and, if 6PE allows for a partial, then it is always beneficial.

Once these design decisions have been made with regards to the core, and a network design completed, then deployment is a matter of planning and implementation. If a trial network was initially built, then the operational management of the IPv6 portion of the network is already well understood. A migration plan needs to exist to include the trial network in the production IPv6 setup.

Global Crossing already had a Global MPLS enabled IP core. It therefore made the decision to use 6PE and just dual stack the access layer. After the design and planning had been completed, the migration from the IPinIP trial setup to all access routers supporting dual stack with 6PE in the backbone took just six weeks, again with no adverse impact its customer base, except through routine and planned maintenance.

There is no right or wrong way to add IPv6 to an existing network infrastructure, and the final design may include elements of 6PE, Dual Stack and IPinIP tunneling. Global Crossing’s IPv6 implementation, although based on 6PE in the core and dual stack on the edge access routers, has also maintained an element of IPinIP tunneling. The original trail devices have remained in the network continuing to support customers that require an IPinIP access mechanism for their network. The tunnel devices are also participating in the 6PE and dual core infrastructure providing a seamless single IPv6 infrastructure.

For further information regarding IPv6 Network Migration, please contact Brian Gottbetter, Dir. of IP Product Management, at brian.gottbetter@globalcrossing.com.