January 1999
Quality Of Service:
Traffic Management On The IP Network
BY ELIZABETH RACIOPPI
There is little doubt that sending voice over the public IP network can exponentially
cut telephony costs. Additionally, VoIP, especially when integrated with video and
multimedia, offers some of the most compelling and creative applications available to
consumers and businesses today. Yet behind this promise of innovation and flexibility
looms the network manager’s nightmare: how to balance time-sensitive (isochronous)
voice traffic with data traffic so that quality and integrity are maintained. When it
comes to audio quality, the bar is set high. The number of companies dedicated to
improving voice quality on wireless networks is a testament to voice quality’s
importance. This imperative for call quality has tremendous implications for the success
of integrated voice applications on the network, and this impact will be felt by both
enterprises and service providers managing the networks that this voice traffic traverses.
To ensure consistent call quality, it is essential to implement a Quality of Service
(QoS) strategy that takes into consideration factors such as classification of traffic,
congestion avoidance, and perhaps most importantly, minimization of latency. To
accommodate these needs, a number of critical factors must be considered. Network managers
must be able to segment delay-sensitive voice traffic while ensuring that bandwidth and
priorities are in line with the applications being run. Policies should also be developed
to identify users and applications making use of VoIP technology. Finally, provisions must
be made for monitoring performance within the network and through a service provider
network, and for tracking Service Level Agreements (SLAs). Instituting a comprehensive QoS
strategy that supports VoIP involves implementing techniques that guarantee the efficient
utilization of bandwidth — namely, traffic management and policy enabled networking.
TRAFFIC MANAGEMENT AND QoS
Until now, some networks managers have attempted to avoid congestion by over-provisioning
bandwidth to support mission-critical applications. This, of course, is expensive —
and by definition, inefficient. In addition, throwing bandwidth at voice traffic does not
alleviate the dreaded latency problems that occur in multi-user networks. But by
incorporating traffic management into QoS strategies, the traffic in both service provider
and enterprise networks can be managed in such a way that network bandwidth is provisioned
properly and cost effectively. QoS can be applied in two ways: to the user or to entire
flows, which consist of aggregated traffic relative to applications or user groups.
Clearly, applying QoS to specific users is the most granular and can be used in an
enterprise environment, while applying QoS to a flow at the edge of a network is easier to
manage by service providers.
Integrated Services
Let’s say a user is running an interactive audio/visual collaboration application
with a key business partner. The user requests additional bandwidth by signaling the
network using RSVP (Resource Reservation Protocol), a signaling protocol. The request is
carried through the networking infrastructure and, based on available bandwidth, will be
granted or rejected. This process is known as IP Integrated Services (IntServ), which
provides a strictly controlled, point-to-point, managed service between end stations.
IntServ provides a guaranteed level of service complete with the controls over delay
that are crucial to voice applications. However, it requires a network path in which all
infrastructure devices are RSVP enabled. If one of the hops along the designated path does
not support RSVP, service will not be guaranteed. In addition, if the state of the network
changes, rerouted packets will not be guaranteed the same QoS. Therefore, IntServ is
appropriate for enterprise networks, but does not scale well in the wide area.
Differentiated Services
Another way of prioritizing bandwidth is to grant levels of service directly to an
application. In this example, a company is running an important ERP (Enterprise Resource
Planning) application such as Baan. The entire application and all of its users are
granted sufficient bandwidth, as identified through Type of Service (ToS) bits in the IP
header. These ToS bits (now known as the DS field) can be read and manipulated by layer 3
devices, placed in corresponding priority queues, and forwarded throughout the network.
This QoS method is known as Differentiated Services, or DiffServ.
DiffServ marked packets carry their own state with them, obviating the need for a
maintained state in the network infrastructure. Service providers can also provide varying
levels of service to voice customers on a less granular scale by prioritizing entire
flows. As an example, a service provider supplying networked services to a company may
provide the entire finance department with premium service at the close of a fiscal
quarter. This method also uses the DiffServ model.
Conditioning
To manage voice traffic using these QoS methods, it is necessary to perform several
traffic management functions, called “conditioning,” at the ingress of the
enterprise network. Conditioning involves several functions that must take place before
forwarding throughout the network occurs: classifying, policing, marking, and shaping.
Classifiers identify traffic relative to predetermined priorities found in both DS bytes
(TOS octet for IPv4) and packet headers (RSVP-type). Policers enforce rate compliance by
using a “token bucket,” which discards packets that exceed an agreed upon rate.
Markers set the DS byte based on the classification, and shapers ensure that aggregate
flows conform and that they do not burst into noncompliance.
If voice traffic is marked as premium, it will be shaped at the edge routers of the
network. Shaping creates regular traffic patterns, as opposed to bursty traffic, by
setting up thresholds that, when exceeded, cause reprioritization of all traffic and may
dictate the designation of a new clean path for the sensitive voice traffic.
The traffic management and QoS methods described above must be supported in routers,
switches, and other infrastructure devices. These parameters must be set manually or by
the application. Each case given above illustrates the selective use of existing bandwidth
as a way of managing resources and users economically. But wouldn’t it be nice if the
network could automatically sense a user’s bandwidth demands and allocate network
resources accordingly? While truly clairvoyant networks are yet to be devised, today we
have the next best thing: policy enabled networks.
“NAME, RANK, AND SERIAL NUMBER” — POLICY ENABLED NETWORKING
Policy management refers to the dynamic allocation of necessary bandwidth and network
resources based upon predetermined criteria. Such policies can give priority to designated
users, groups, or applications based upon the state of the network and available bandwidth
at a given time. By utilizing bandwidth only where and when it is needed, policy enabled
networking can dramatically increase cost savings and performance. This method of
provisioning QoS in IP networks relies heavily on maintaining user profiles, as well as
monitoring network status to ensure minimal congestion.
User profiles consist of information about the applications individuals are authorized
to use, when they’re authorized to use them, what security privileges they have, what
areas on the network they have access to, and what their QoS requirements are. Profiles
are determined by a combination of users, service providers, and network managers, and are
maintained in directories. It is best that these directories be accessed through a
standard access protocol, such as LDAP (Lightweight Directory Access Protocol). Profiles
can be built and stored in directories located throughout the network rather than in one
centralized location.
Thus, user profiles can be proliferated within and across networks so that access
rights and privileges can be tracked and managed easily. Policy enabled networking makes
use of policy managers or enforcement points: i.e., decision points within the network
that communicate policy information to network devices. Communication between the policy
manager and the enforcement points (routers, switches, and the like) is done via
bi-directional protocols. The most visible emerging policy protocol today is called COPS
(Common Open Policy Service), which is currently on its way to becoming a standard through
the IETF.
Policy enabled networking makes the provisioning of QoS and other network resources a
dynamic process, resulting in the most significant benefits of policy enabled networking:
economy and performance. These efficiencies will allow IP intranets and extranets to meet
the demands of converged voice, video, and data.
PARDON ME, MAXWELL, IS THAT YOUR SHOE-PHONE RINGING?
Have you noticed that voice applications are showing up in the strangest places these
days? Take your PC, for instance. Five years ago, who would have thought that talking to
your PC for such things as video conferencing and e-commerce would be a commonplace
alternative to picking up the telephone?
As applications become more sophisticated and incorporate more of the voice/video/data
mix, technologies such as Internet telephony become more compelling. Suppliers of
traditional phone service are considering putting voice packets on the Internet as a less
expensive alternative to the PSTN. Digitizing these voice packets and putting them on an
IP network means that these packets are now sharing overall network bandwidth, rather than
requiring more costly point-to-point connections. From an administrative standpoint, voice
on a network is managed in the same management domain as other data types, setting the
stage for an economy of scale.
Additionally, voice over IP technology can facilitate innovative solutions such as
Web-based call centers, and even “virtual call centers” that are location
independent. With agents located at any remote facility with a standard Internet
connection, the overhead of maintaining facilities is minimized. However, voice is
sensitive traffic and must be treated differently to minimize latency.
The Internet will be the preferred network transport for many years to come. There is
no reason for any user, organization, or company to settle for a best-effort network to
support their voice traffic. By developing a QoS strategy and implementing a
policy-enabled networking architecture, users and businesses can feel confident that their
voice-based and mission-critical applications will be managed as though they were using a
private network. They will see that a higher quality of networking can be achieved, costs
can be managed, and innovative solutions can be implemented that will revolutionize the
way we communicate in this new era of networking.
Elizabeth Racioppi is the senior technology marketing manager for Nortel Networks.
Nortel Networks works with customers worldwide to design, build, and deliver
telecommunications and IP-optimized networks. Customers include public and private
enterprises and institutions; Internet service providers; local, long-distance, cellular,
and PCS communications companies; cable television carriers; and utilities. For more
information, contact them at 800-4-NORTEL, or visit their Web site at www.nortelnetworks.com.
|
TMCnet LOGIN
Webinars
