Employing Quality of Service
The previous section presented two
means of examining the network to gather a fundamental understanding of how the
network is being used. This information forms the foundation from which you can
make decisions about how to align network resources to accommodate the relative
priority of applications and hosts that are using the network. This information
is also important to ensure that the network is configured in such a way that it
provides the appropriate levels of handling and control based on application and
business requirements.
Using information gathered through
NetFlow or other mechanisms (for instance, network analysis modules or
accelerators with monitoring capabilities, both of which generally provide
similar reporting capabilities), network administrators can then define what
levels of service need to be applied to different types of traffic and how this
data should be handled. These same tools also serve as a means of validation
after such policies have been implemented within the network to verify that the
application requirements and business priority definitions are being met.
Without QoS, best-effort handling is
provided to all flows on the network, thereby introducing the possibility that
noncritical traffic may delay or block the service of business-critical traffic
or traffic from applications that have sensitivities to network characteristics
such as latency or loss. With best-effort handling, all traffic is considered
equal, as shown in Figure 3-7. In this figure, peer-to-peer sharing consumes all
the available network resources, leading to bandwidth starvation for the
applications that need precious WAN bandwidth to drive user productivity. Such
situations could quickly lead to employee frustration with business
applications, loss of revenue, poor morale, and job dissatisfaction.
With QoS configured, appropriate
handling can be provided to traffic flows based on classification and priority.
As shown in Figure 3-8, the result is that the business-critical applications
that demand bandwidth and service on the network are unimpeded even when
unsanctioned traffic is present. In this example, peer-to-peer sharing is shown as
being permitted, and it should be noted that such applications could be blocked
altogether.
Note
Figure 3-8 shows peer-to-peer sharing as permitted, however, this
traffic could be blocked rather than permitted.
With QoS in place, network
administrators are able to better allocate available network capacity to those
applications that need it the most. This allocation of network resources with
business priority helps to ensure precious network capacity is used by
sanctioned applications, thereby encouraging user productivity.
The QoS architecture is built
around a behavioral model that comprises the following four key functions, each
of which is outlined in the following sections. These functions provide the
facilities necessary to align network resources with business priority and
application requirements.
-
Classification: Identifies application and traffic flows on the
network. Once identified, further action and specific handling can be applied to
the flow.
-
Pre-queuing: Includes operations that are performed
against flows prior to consuming network device (router or switch) resources
such as queues. The operations include dropping packets (undesirable flows),
traffic conditioning (policing), and marking relative priority on the packets
themselves.
-
Queuing and scheduling: Enforce priority of selected packet streams through the use
of configurable queuing mechanisms, such as high-priority handling of
delay-sensitive traffic, selective delay of lower-priority traffic during
periods of congestion, traffic conditioning (shaping), and enforcement of
bandwidth allocation.
-
Post-queuing: Improves link throughput using optional
operations such as packet compression, header compression, and link
fragmentation and interleaving (LFI).
The behavioral model provides the
facilities necessary to align network resources to business priority and to
optimize user productivity over the network. Figure 3-9 shows an example.