QoS Capability

Table A-1 provides a fictitious allocation of traffic types to packet-marking values. Each enterprise is likely to have a different scheme; this table is provided for illustrative purposes only.

Table A-1. QoS Classes of Service

CoS/IP Precedence	Forwarding Method	Description
6	CBWFQ min-bw	Network control and management
5	LLQ	Voice bearer traffic
4	CBWFQ min-bw	Videoconferencing bearer traffic
3	CBWFQ min-bw	Call signaling and high-priority data applications
2	CBWFQ min-bw	Interactive traffic
1	CBWFQ min-bw	"Scavenger"Bulk traffic dropped first during congestion
0	FBWFQ	Default classAll other traffic

The provider doesn't need to match these classes exactly. However, the provider must provide a number of classes that match or exceed the required QoS of each of these traffic types and be able to map these markings to those classes transparently. Assuming that the provider offers three classes, the lowest-priority CoS typically needs a 99.9 percent packet delivery rate, as detailed in Table A-2.

Table A-2. Required Service Provider Classes of Service

CoS/IP Precedence	Forwarding Method	Packet Delivery Rate
Low-latency/priority queue	LLQ	99.995 percent
High-priority data	CBWFQ	99.990 percent
Normal-priority data	CBWFQ	99.900 percent

To ensure that the enterprise can recover and revert to its own QoS markings for its seven classes at the opposite end of the IP/VPN service, it is a requirement for "subclasses" within each primary CoS whose IP Differentiated Services Code Point (DSCP) values remain consistent upon arrival at the remote CE node. For example, a service provider may offer three CoS levels. These may be "silver" and "bronze" based on IP DiffServ's "assured forwarding" per-hop behavior and "gold" based on IP DiffServ's "expedited forwarding" per-hop behavior. There must be a mechanism by which the enterprise can map its CoS values 6, 4, 3, and 2 (from Table A-1) into the "silver" CoS (perhaps as IP DSCP values CS4, AF41, AF42, and AF43) and have them retained such that this mapping can be reversed into the original seven CoS values for the enterprise's network. In provider networks where this capability is not present, complex reclassification for traffic reentering the enterprise network from the provider network is necessary.

Low-speed links may need to use Link Fragmentation and Interleaving (LFI) to support the enterprise's end-to-end jitter requirements. Low speed is typically defined as less than 768 kbps. Typically, it is required that jitter as measured from CE to CE be no more than 30 ms, based on the capability of end device jitter buffers to compensate for this effect. Service providers are required to have the capability to use LFI for low-speed links (1 Mbps and less) to meet this jitter requirement.

These issues can be captured in the following questions:

• Number of classes of service How many classes of service are available for use within your IP/VPN service, and which type of service is offered with each class (low-latency queuing, minimum-bandwidth, or best-effort)?

• QoS transparency Is your network configured to preserve the IP DSCP values sent from the CE router throughout your network until it reaches the remote CE router?

• Number of QoS "subclasses" that can be used in any given class of service How many different "subclasses" or DSCP markings can be used in each of your classes of service, and what (if any) special handling is applied (such as greater drop precedence)?

• Traffic engineering Do you perform any traffic engineering, which may place traffic in a lower CoS on a different path than the highest CoS?

• LFI Do you provide LFI capability on the PE router for all low-speed links? If so, please explain the type(s) of LFI used for each access method where committed bandwidth would be 1 Mbps or less.

• cRTP What support for Compressed Real-Time Protocol (cRTP) is available, and in what locations?