|
In the new
infrastructure, IP must provide business-quality solutions.
The key to
providing QoS in the new service architectures is the ability to
differentiate traffic and to provide differentiated service levels
based on the types of traffic. For example, for real-time
applications such as voice over IP (VoIP), the amount of available
bandwidth and end-to-end delay is crucial compared to fax and e-mail
transmissions, which are quite insensitive to bandwidth and delay
issues.
To provide QoS
from a network, the following must be satisfied:
-
User/application requirements should be known to the network; and
-
The network
should have appropriate mechanisms for providing service levels
that approximate these requirements.
The standard IP
architecture was never designed to deliver on either of the two; it
is based on a “best-effort” model where all network traffic is
equally important and everyone receives service based on
availability, without guarantees.
In the absence of
QoS mechanisms, the industry traditionally has opted for
over-provisioning bandwidth. While bandwidth over-provisioning
continues, industry experts agree that QoS mechanisms are needed to
address the needs of converging networks.
In general,
end-to-end QoS in Internet is built from the concatenation of
edge-to-edge QoS from each domain through which traffic passes, and
ultimately depends on the QoS characteristics of the individual hops
along any given route. The solution can be broken into three parts:
-
per-hop QoS,
-
traffic
engineering, and
-
signaling/provisioning.
QoS mechanisms do
not generate more bandwidth. They manipulate router/switch queues so
that when congestion occurs, priority “VIP” traffic is serviced
quickly, while less important traffic experiences delays and drops.
The network applies packet-filtering criteria to identify and
prioritize VIP traffic using either provisioned or signaled QoS:
Provisioning assumes the network nodes are configured ahead of time,
while signaling assumes that filtering criteria are communicated in
real time, upon demand. Classification is based on class of service
(CoS) or QoS criteria. QoS deals with individual flows; CoS,
generally considered a subset of QoS, deals with aggregate classes
of traffic. After being classified, packets are serviced by a
predefined queuing discipline that determines their final service
level.
Because not all
QoS mechanisms are the same, selecting the right QoS mechanism for
the network could affect results significantly.
Figure 3 -
Per-hop classification, queuing, and scheduling (CQS) routing
architecture
A variety of
standard QoS approaches are common. Some router/switches are capable
of setting filters to classify traffic and map it to specific
queues. Some of the more popular disciplines include:
-
Priority
Queuing (PQ),
-
Class-Based
Queuing (CBQ),
-
Weighted Fair
Queuing (WFQ), and
-
Random Early
Detection (RED).
In all these
approaches, the entire QoS process (classification and queuing) is
provisioned within a single node and requires no cooperation from
others. However, the process of filtering packets based on multiple
attributes causes
-
High overhead,
-
Does not scale
well, and
-
Hard to create
consistent multi-hop QoS by preconfiguring individual routers in a
routing insensitive manner.
Figure 4 - A
simplified Native IP Forwarding (NIF) engine
An important
issue in the Internet, and consequently in every network connected
to it, is support for multimedia applications (video, voice). These
applications have specific requirements in terms of delay and
bandwidth which challenge the original design goals of IP's best
effort service model, and call for alternate service models and
traffic management schemes that can offer the required quality of
service (QoS). To this end, two QoS architectures have emerged in
the IETF:
·
Integrated services architecture
(IntServ), which provides end-to-end QoS on a per-flow basis;
features soft states and end-to-end signaling.
·
Differentiated services architecture
(DiffServ), which supports QoS for traffic aggregates; features
class of flows and code points contained in the IP header’s
differentiated services field.
Both proposals
suggest solutions to overcome the QoS limitations in the current
best-effort IP service architecture. Each system has, however, its
own advantages and disadvantages, and its own role to perform in an
appropriate segment of an IP network.
We now review
these two proposals on how such QoS enabling schemes could be
utilized to enhance the best effort service model of IP architecture
|
