After several years of attempting to differentiate and
classify services using network-based Quality of Service
(QoS) technology, the telecommunications industry is
waking up to a new business reality. Service providers
have learned that profitably delivering new services,
such as IP telephony and hosted applications, requires
more than packet-based QoS solutions that emphasize
router-to-router performance. Service providers need
solutions that can help them assure delivery of
applications and services all the way to the end
subscriber.
Service providers are also actively looking for ways
to profitably deliver lucrative new services that
increase average revenue per user (ARPU), without
requiring the replacement of existing infrastructure.
Simply overbuilding the network has proven to be
impractical and unprofitable. And attempts at packet
classification, with their costly infrastructure changes
and excessive management complexity, have not delivered
convincing results. Providers and their customers are
asking for solutions that can scale to support tens of
thousands of subscribers, do not require hundreds of
man-hours to implement and maintain, and provide
effective service delivery without disturbing their
existing infrastructure.
Application Prioritization -- Starting With The
Subscriber
Today, providers are evaluating a new approach that
prioritizes and controls applications and services
precisely where they are used -- at the subscriber end
station. This approach differs from QoS because it
focuses on aligning network resources directly with
subscriber demands and revenue opportunities. By
implementing control and prioritization at the end user
device (e.g., PC, palmtop, IP phone), providers can now
effectively and profitably address the distributed
nature of subscribers and applications.
This distributed approach applies and enforces
business-driven policies at the session layer (layer 5)
of the OSI model, between the application and the
network. Session layer implementation provides the
ability to "see" traffic flows in terms of the
originating application or service. Policies are
therefore targeted at high-level entities such as users,
applications, and groups. This approach makes management
a much simpler task, as policies are based on intuitive
business concepts. The need to deal with messy packet
definitions or router configurations is eliminated. In
addition, the billing of services based on usage becomes
much easier, because a direct correlation exists between
the policy and the entity or service being billed.
Solutions based on session layer prioritization use
an architecture consisting of centralized policy
management and distributed policy enforcement. Policies
based on users, applications, and services are managed
centrally at a policy management engine. These policies
are distributed to agents at the end devices for local
session layer enforcement. This unique design provides
holistic visibility into users, applications, and
resources throughout the network, while leveraging the
power and performance of end devices.
Getting Results With Existing Networks
Among the biggest problems of QoS has been the need for
complex implementations. QoS technologies such as
Differentiated Services (DiffServ), Multi-protocol Label
Switching (MPLS), Resource Reservation Protocol (RSVP),
and proprietary QoS appliances require router updates or
replacements, or the addition of new gear. Managers
therefore must visit and reconfigure virtually every
WAN, which is an effort that is simply impractical on
most large networks. In many cases, applications must
also be upgraded or modified to take advantage of the
QoS capabilities of the new equipment. And managers must
be diligent about interoperability among all QoS
devices, which are extremely sensitive to firmware
revisions.
Application prioritization resolves this dilemma by
removing the need to touch the existing network
infrastructure. These new systems transparently overlay
onto networks with no configuration requirements or
considerations. Managers can deploy these solutions even
as their networks continue to run at full capacity.
Desktop agents are distributed to client devices through
Web-based installations or management systems such as
Microsoft SMS, and can be rolled out at a pace to meet
any schedule.
The Scalability Challenge
Another challenge for traditional QoS is scalability.
Standard QoS solutions require that traffic reach a
central router or QoS appliance before any control or
shaping takes place. As the network grows, or traffic
loads increase, a classic bottleneck occurs as too many
packets attempt to pass through the now overloaded QoS
device. Once the bottleneck occurs, packets are either
queued for delayed delivery, or dropped completely,
forcing costly retransmissions. This is a bad outcome
for all users, and particularly negative for
delay-sensitive applications, such as VoIP and streaming
media.
The focus on application prioritization offers an
important breakthrough for implementing truly scalable
service delivery. The enforcement agents efficiently use
the power of each desktop CPU to carry out the service
delivery workload. This means that as the network grows
to tens of thousands of users, additional resources
become available to help ensure reliable delivery for
everyone. The agents also help enhance the performance
of existing routers, by moving the prioritization
workload off the network, and allowing routers to focus
on their primary task of routing.
The session layer control of application
prioritization has another positive effect on
scalability -- the elimination of costly congestion. A "hidden"
drawback to traditional QoS has been the lack of
congestion control. Because traffic runs unrestrained
until it reaches the first QoS device, congestion is a
permanent artifact of the network. With the session
layer approach to prioritization, traffic is sorted and
prioritized before it ever enters the network. This
results in smoother and more predictable delivery for
all applications and users, regardless of their
priority, and keeps bandwidth working at maximum
efficiency.
Focusing On Service Revenue And Profitability
In addition to scalability and greater efficiency,
application prioritization can directly generate new
revenue for providers. For example, VoIP service is a
powerful companion offering to business customers
signing up for VPNs. After all, why shouldn't customers
leverage their VPN investment to reduce their voice
costs? This has proven to have been an awkward sales
strategy, because providers have had no way to ensure
the delivery of time-sensitive VoIP on VPNs. The process
of data encryption prevents the effectiveness of
traditional QoS. The new prioritization systems control
VPN traffic at the session layer, prior to data
encryption. So, providers can now offer VoIP as an
add-on sale to VPN customers with the confidence that
they can ensure adequate delivery.
Distributed service prioritization offers value well
beyond ensuring and controlling delivery. With services
initiated right at the subscriber's device, providers
can deliver new capabilities that go well beyond the
traditional bounds of QoS. For example, providers can
now include such services as billing by application or
usage, built-in level one customer support, desktop
provisioning, and encryption services. Simply
distributing the agent to subscriber desktops enables
each of these add-on services. The add-on service then
resides quietly at the desktop until the provider is
ready to put the capability into action.
Conclusion
Distributed service prioritization provides a profitable
new focus on the application and subscriber. With its
non-invasive implementation and distributed scalability,
this new approach helps providers ensure the efficiency
of their networks and reliable delivery for customers,
while expanding revenue opportunities with a new
generation of services.
Suketu Pandya is co-founder and CTO of Centricity
Software. Centricity Software provides solutions
that uniquely enable distributed service providers and
enterprises to deliver and differentiate application
services to their users, partners, and customers.
[ Return
To The July 2001 Table Of Contents ]
|