Deploying QoS for IPv6

QoS is always enabled in an environment that already provides unicast connectivity. The tools and features used to enable this service depend on the characteristics of this infrastructure. IPv6 most likely is the newcomer in a preexistent IPv4 network. Chapter 3 discussed the IPv6 unicast deployment options and how the QoS design has to be tailored to them. This section looks at design recommendations for native and MPLS-based IPv6 deployments. Inevitably, the coexistence of the two protocols has to be addressed, too.

QoS in a Native IPv6 Deployment

The similarity between the implementations of QoS for the two versions of the IP protocol implies that the IPv4 QoS deployment rules are applicable to IPv6, too. You can find a detailed analysis of these design rules in the book End-to-End QoS Network Design: Quality of Service in LANs, WANs, and VPNs, by Tim Szigeti and Christina Hattingh. The basic steps to follow in planning for and designing a QoS deployment are as follows:

1.	The most important first step that must be taken before designing or even considering the deployment of QoS is that of defining its business objectives network. A DiffServ option is discussed here for the QoS deployment. Note IntServ might be considered for applications such as videoconferencing that benefit from reserving resources prior to sending the traffic. The IntServ implementation most likely would be an overlay on top of DiffServ used for most traffic types.
2.	A thorough evaluation of the network capacity is required. The identified constraints imposed by the existent infrastructure provide the framework for prioritizing the business objectives and the resources to be allocated to them.
3.	The outcome of the first two steps is the set of classes that are used throughout the network to differentiate the traffic types. Three to five classes typically suffice for most deployments; too many classes make the service difficult to manage.
4.	Classify the traffic as close as possible to the source and mark it using DSCP.
5.	Define the PHBs for the various devices in the network. The devices that can perform the QoS functions in hardware should be more heavily used in the design. Policing should be performed as close to the source as possible. Note QoS classification and marking can and should be done prior to it being encrypted and forwarded through tunnels.
6.	Test and verify the QoS design before deployment.

The network layers usually support the end-to-end QoS in different ways:

• Access This layer provides the best opportunity to enforce customer traffic profiles through policing and rate limiting. This layer is also the closest to the sources of traffic and it represents the best place to classify and mark packets. The access layer might use more classes than other layers of the network to sort more finely customer traffic. The QoS function can be done in both layer 2 and layer 3 devices.

• Edge/aggregation The traffic that reaches the edge/aggregation layer is typically marked and conditioned by the access layer. At this point, the traffic might be policed again, but it will be shaped before entering the core. The number of classes is reduced to the recommended three to five, and hence the packets might be remarked.

• Core In the past, a common approach was to rely on a high-bandwidth core that would be engineered so that it would not see congestion (thus rendering the use of QoS unnecessary). Even today, some argue that in SP environments that it is cheaper to oversize the core network than to design it with QoS. When network bandwidth starts to be scarce, it is more convenient to increase link capacity than to deploy any sort of mechanism that brings complexity and maintenance overhead. Making the core aware of application, protocols layers (for instance, IPv4 and IPv6), customers, and so on may not be considered as a scalable thus viable solution. In such environments, only edge policies will be configured. However, this general philosophy may turn to be expensive for a number of reasons, including the following:

  - The network must accommodate peaks, sometimes an order of magnitude more than the average traffic. Turning on QoS in the network to differentiate traffic may prove to be more economic than oversizing the network.

  - The network must be able to react well to link failures, where traffic is rerouted on alternate paths. When the failure occurs on a large trunk, and/or when the failure affects several trunks, some congestion can build on smaller links, which were not sized for absorbing the rerouted traffic.

  - Denial-of-service attacks can always saturate links. QoS allows some time-sensitive traffic such as voice to work perfectly even at the peak of a denial-of-service incident.

  - The implementation of QoS in the core is probably the simplest. The traffic is already conditioned by the edge/aggregation and only a few (equal to the number used in the edge/aggregation layer) classes are used.

Similar to IPv4, congestion control and management are implemented depending on the characteristics of the traffic identified through the various classes.

Layer 2 QoS should be leveraged as much as possible throughout the network because switches often implement it in hardware. Usually a mapping of the top three DSCP bits or the TOS bits to COS is sufficient in terms of marking. However, the option is available to have independent marking policies.

QoS in an MPLS-Based IPv6 Deployment

The SLA and its associated network-deployed enforcing mechanism (QoS) are end-to-end concepts. Therefore, it is logical that it is implemented primarily by service providers, who are the ones with the end-to-end perspective. Because MPLS is one of the dominant technologies deployed in SP networks, it is also one of the primary areas for QoS implementation.

Similar to native IP QoS, MPLS QoS technologies can be split into IntServ and DiffServ. The latter is the most widely deployed mechanism, but one starts to see some deployments with MPLS traffic-engineered tunnels, to provide guaranteed bandwidth across the MPLS core. MPLS-TE uses RSVP-TE to set up a labeled path across the MPLS domain. MPLS DiffServ and MPLS-TE can be combined together into an even more powerful tool for delivering QoS on packet networks. Most of the MPLS QoS deployments focus on the network edge, although some service providers are starting to deploy QoS in the MPLS core, too.

Both 6PE (IPv6 MPLS over v4-signalled Label Switch Path, described in Chapter 3) and 6 VPE (BGP MPLS IPv6 VPN over v4-signalled LSP, described in Chapter 7) involve dual-stack edge routers communicating over a v4-based MPLS backbone. To enable QoS for 6PE and 6VPE, IPv6 traffic must be classified and conditioned (policed, shaped, and marked) pretty much the same way as IPv4, before entering the MPLS backbone. This typically must take place at the customer edge (CE) routers in the case of a managed service or at the provider edge (PE) router otherwise. This implies identifying IPv6 classes, then performing policing, shaping, and marking. In Figure 5-5, those tasks are referred to as edge policies. In practice, either the CE outbound policy or the PE inbound policy is set up: CE inbound policy is never used. In addition, in case QoS is also implemented in the MPLS core, a PE outbound policy is configured together with outbound policies at each hop (P-routers) in the MPLS provider core. This is the PHB, which, in this case, is essentially queuing and dropping.

Figure 5-5. Deploying IPv6 QoS over MPLS

MPLS is the foundation of a multiservice network, which is intended to transport a large variety of network layer protocols (IPv4 unicast and multicast, IPv6 unicast and multicast, IPv4 VPN, IPv6 VPN, as well as ATM, FR, PPP, Ethernet, and so on) and a variety of application data (Internet content, VoIP, video, and so on). When deploying QoS in the MPLS core, one does not expect to make it aware of the network layer protocol or of the application being carried. In other words, it is unlikely and even not recommended to differentiate IPv6 traffic from IPv4, but rather to treat equally IPv4 and IPv6 real-time traffic, and differentiate them from IPv4 and IPv6 data traffic, for instance.

In that context, enforced by the 6PE (and 6VPE) approach where the same LSP used to transport IPv4 and VPNv4 traffic is also used to transport IPv6 traffic, the setting of core QoS for the latter is straightforward. It is just a matter of classifying the 6PE/IPv6 traffic into existing MPLS classes, and marking MPLS-encapsulated IPv6 traffic accordingly.

Note

There is no difference regarding DiffServ in an MPLS-based IPv6 deployment, whether the LSP was set up using IPv4 (6PE and 6VPE) or IPv6 or whether it was setup using LDP, RSVP, or even BGP.

Instead of, or in addition to DiffServ, there may be some interest in deploying an IntServ strategy across the core (for instance, to reserve some paths for specific customers, or to optimize a better use of the network bandwidth). This can be done using MPLS-TE.

MPLS, combined with RSVP, provides a mechanism to set up explicit paths across the core, and to associate them with bandwidth reservation or even DiffServ strategies. Despite the management complexity, some service providers with an MPLS infrastructure have meshed their core with TE tunnels and manually set up paths with reserved bandwidth. In that context, DiffServ can also be used to manage congestion conditions inside the TE tunnels themselves.

When using MPLS-TE in networks where IPv4 and IPv6 traffic coexist, two approaches are possible: Either TE tunnels are shared for IPv4 and IPv6 traffic (recommended) or separate tunnels are signaled and used for the two protocol versions. TE tunnels can be combined with DiffServ, where the selection of tunnels is performed based on DSCP bits. This is referred to as CBTS. DiffServ-aware traffic engineering enables service providers to perform separate admission control and separate route computation for discrete subsets of traffic (for example, voice and data traffic, regardless of the network layer protocol).

Note

When IPv6-RSVP is available, it might still make sense to use a single MPLS-TE setup mechanism (IPv4 or IPv6) to set up a single tunnel used by both IPv4 and IPv6 for certain traffic classes.

IPv4 and IPv6 Coexistence

One question arises with the coexistence of IPv4 and IPv6: Should IPv6 traffic be differentiated from IPv4 traffic, or rather should traffic of a given class (for instance, real time) be differentiated from traffic of other classes (for instance, data) regardless of IP protocol type? Not only are both approaches possible, but different strategies can also apply to different parts of the network.

The PHBs for the two protocols might be different under the following considerations:

• IPv4 traffic is revenue generating, and it most likely is more important for the business than the IPv6 traffic, at least in the beginning. In that case one might choose to prioritize the IPv6 traffic lower than IPv4 and provide fewer resources for it.

• IPv4 and IPv6 traffic might have to observe different PHBs depending on traffic patterns used by various applications.

In these cases, you should define different classes and policies for each traffic type.

Note

With transition mechanisms, the IPv6 traffic can leverage the deployed QoS of the traversed IPv4 infrastructure. In some circumstances, the IPv6 traffic might also lose its markings after crossing the IPv4 network.

Differentiating based on applications rather than network layer protocol makes the most sense: After all, end users running a particular application (for instance, IP telephony) do not care about the network layer being used and expect the same level of network quality in all cases. This approach also reduces the management overhead for the QoS deployment and the use of network element resources.

It is recommended that no differentiation should be made between the two protocol types in implementing QoS at layer 2. Similar to the recommendation made for MPLS, the infrastructure should be kept unaware of the transported protocol type. Moreover, different PHBs for IPv4 and IPv6 would lead to an unmanageable number of policies.

The aim of IPv6 to reestablish a peer-to-peer model for IP transport will most definitely impact in a positive way the deployment of QoS. The boundary between private and public domains will no longer even out the characteristics of individual streams coming from different internal sources. In an IPv6 world, true end-to-end QoS policy implementations are closer to reality. Despite lacking a consolidated architecture at this time, the features available for IPv6 do provide the means to deploy QoS at least at the level of current IPv4 deployments.