Advanced QoS Mechanisms in the Network
Classification
Policing/Shaping
Link Efficiency
256k 128k
Frame Relay
Congestion
Management/
Avoidance
Hub Site
Policing/Shaping 384k
Congestion
Management/
Avoidance
Congestion
Management/
Avoidance
Congestion
Management/
Avoidance
T1 128k
Classification
Classification
Link Efficiency
Apps Signaling Apps
By classifying packets at the edge of the network into distinct classes, the network
can provide differential services to the packets without having to examine
each one in detail at every hop. After they are marked once with IP precedence,
congestion management and avoidance mechanisms can act upon them as they
travel to their destination.This is the essence of the DiffServ model of QoS. Since
DiffServ does not employ end-to-end signaling between clients, it is basically a
connectionless form of QoS. Although it may not be able to guarantee QoS totally,
it will scale well and make efficient use of spare network resources.
www.syngress.com
Figure 8.9 Advanced QoS Mechanisms in the Network
Classification
Policing/Shaping
Link Efficiency
256k 128k
Frame Relay
Congestion
Management/
Avoidance
Hub Site
Policing/Shaping 384k
Congestion
Management/
Avoidance
Congestion
Management/
Avoidance
Congestion
Management/
Avoidance
T1 128k
Classification
Classification
Link Efficiency
Apps Signaling Apps
272 Chapter 8 • Advanced QoS for AVVID Environments
On the other hand, we also looked at the Intserv model embodied by RSVP.
Since RSVP is a signaling protocol between clients, it can be thought of as connection-
oriented QoS model. Connection-oriented networks are traditionally good
at providing QoS guarantees, but they do not make efficient use of spare bandwidth,
and they have serious scaling problems.
So which approach should you take? There is no doubt that for most networks,
DiffServ and IP precedence will be convenient and functional. Cisco’s
direction seems to indicate a growing trend towards creating and improving
mechanisms that will heed class-marking bits such as IP precedence and
Differentiated Services Code Point (DSCP).The lack of RSVP-enabled clients
and other Intserv signaling mechanisms also points towards DiffServ. However, in
future networks, the most effective QoS will probably be found by using a combination
of the two models. One can imagine a large network (maybe even the
Internet) with a DiffServ core, enhanced, perhaps, by technologies such as
Multiprotocol Label Switching (MPLS), with an Intserv function like RSVP
working at the client level.With technology changing as rapidly as it is, it is very
difficult to predict.