Hardware Support for DS-TE LSP
Juniper Networks supports two kinds of DS-TE LSPs: DiffServ-aware single-class LSPs and DiffServ-aware multi-class LSPs. Single-class LSPs are similar to traditional L-LSPs, and support only one class per LSP. Multi-class LSPs can be thought of as L-LSPs that can handle multiple classes. Each multi-class LSP can support up to four classes with specific bandwidth reservation assigned to each class. When DiffServ-aware LSPs are routed on a network, consideration is given to the amount of bandwidth reserved on each interface for each class. If there is insufficient bandwidth on a particular interface for a given class on the multi-class or single-class LSP, the LSP will not be routed over that interface.
Class Type
A class type is a collection of traffic flows that is treated equivalently in a DiffServ domain. A class type maps to a queue and is much like a class-of-service (CoS) forwarding class in concept. It is also known as a traffic class.EXP Bits
EXP Bits
The Experimental bits, or EXP bits, in the MPLS header are used to define the class to which a packet belongs. A unique EXP bit pattern is associated with each class type and forwarding class defined on a DiffServ-aware router.
Forwarding Class
Forwarding classes are defined on each router and assigned to internal queues. The default forwarding classes are: best-effort, expedited-forwarding, assured-forwarding, and network-control. Individual class types in DiffServ-aware LSPs are mapped to individual forwarding classes at the router. The default mapping is shown in the table below.
Class Type |
Forwarding Class Name |
---|---|
CT0 |
best-effort |
CT1 |
expedited-forwarding |
CT2 |
assured-forwarding |
CT3 |
network-control |
Scheduler Map
The treatment given to each forwarding class on an interface is defined by the scheduler map assigned to that interface. The scheduler map includes a list of schedulers which map specific forwarding classes to specific scheduler configurations. These determine the per-class bandwidth allocations on each interface, which are taken into consideration when routing DiffServ-aware LSPs.
Bandwidth Model
A bandwidth model must be configured on all routers participating in the DiffServ domain. The three types of bandwidth models supported by Juniper are MAM, Extended MAM, and RDM, which are defined in the following table.
MAM |
Defined in Internet draft draft-ietf-tewg-diff-te-mam-03.txt |
Extended-MAM |
A proprietary bandwidth model that behaves much like standard MAM. If you configure multiclass LSPs, you must configure the extended MAM bandwidth model. |
RDM |
Makes efficient use of bandwidth by allowing the class types to share bandwidth. RDM is defined in Internet draft draft-ietf-tewg-diff-te-russian-05.txt |
Operation
In order to take advantage of DiffServ aware single-class and multi-class LSPs, each class type must be configured consistently across the differentiated service domain. In other words, each router in the network must follow a consistent class type configuration. On each node router, each class type is mapped to a queue.
The available bandwidth for a particular class type on a link is determined by the configuration of class of service queues for that interface. Any DiffServ-aware LSP that requires bandwidth from a particular class cannot be established through routers that do not understand the Classtype object. It is possible for DiffServ-aware LSPs and regular LSPs to be established on the same router. In this case, the regular LSP will carry best-effort traffic by default. However, you cannot simultaneously configure multi-class LSPs and single-class LSPs on the same router.