MPLS Traffic Engineering - DiffServ Aware (DS-TE)

MPLS Traffic Engineering - DiffServ Aware (DS-TE) Background and Overview

MPLS traffic engineering allows constraint-based routing (CBR) of IP traffic. One of the constraints satisfied by CBR is the availability of required bandwidth over a selected path. DiffServ-aware Traffic Engineering extends MPLS traffic engineering to enable you to perform constraint-based routing of "guaranteed" traffic, which satisfies a more restrictive bandwidth constraint than that satisfied by CBR for regular traffic. The more restrictive bandwidth is termed a sub-pool, while the regular TE tunnel bandwidth is called the global pool. (The sub-pool is a portion of the global pool. In the new IETF-Standard, the global pool is called BC0 and the sub-pool is called BC1. These are two of an eventually available eight Class Types). This ability to satisfy a more restrictive bandwidth constraint translates into an ability to achieve higher Quality of Service performance in terms of delay, jitter, or loss for the guaranteed traffic.

For example, DS-TE can be used to ensure that traffic is routed over the network so that, on every link, there is never more than 40 per cent (or any assigned percentage) of the link capacity of guaranteed traffic (for example, voice), while there can be up to 100 per cent of the link capacity of regular traffic. Assuming that QoS mechanisms are also used on every link to queue guaranteed traffic separately from regular traffic, it then becomes possible to enforce separate "overbooking" ratios for guaranteed and regular traffic. In fact, for the guaranteed traffic it becomes possible to enforce no overbooking at all—or even an underbooking—so that very high QoS can be achieved end-to-end for that traffic, even while for the regular traffic a significant overbooking continues to be enforced.

Also, through the ability to enforce a maximum percentage of guaranteed traffic on any link, the network administrator can directly control the end-to-end QoS performance parameters without having to rely on over-engineering or on expected shortest path routing behavior. This is essential for transport of applications that have very high QoS requirements such as real-time voice, virtual IP leased line, and bandwidth trading, where over-engineering cannot be assumed everywhere in the network.

The new IETF-Standard functionality of DS-TE expands the means for allocating constrained bandwidth into two distinct models, called the "Russian Dolls Model" and the "Maximum Allocation Model". They differ from each other as follows:

Table 1 Bandwidth Constraint Model Capabilities

MODEL	Achieves Bandwidth Efficiency	Ensures Isolation across Class Types		Protects against QoS Degradation...
		When Preemption is Not Used	When Preemption is Used	...of the Premium Class Type	...of all other Class Types
Maximum Allocation	Yes	Yes	Yes	Yes	No
Russian Dolls	Yes	No	Yes	Yes	Yes

Therefore in practice, a Network Administrator might prefer to use:

•the Maximum Allocation Model when s/he needs to ensure isolation across all Class Types without having to use pre-emption, and s/he can afford to risk some QoS degradation of Class Types other than the Premium Class.

•the Russian Dolls Model when s/he needs to prevent QoS degradation of all Class Types and can impose pre-emption.