You are on page 1of 4

White Paper

ACCEDIAN Establishing Real-Time Hierarchical


NETWORKS QoS with Zero-Latency Traffic Shaping
PACKET PERFORMANCE ASSURANCE TM

Establishing Real-Time Hierarchical QoS with Zero-Latency Traffic Shaping

The Need for QoS Management


Over-provisioning bandwidth is a possible solution, but offers
In an unconditioned Ethernet or IP/MPLS connection all inconsistent results with the additional cost of consuming core
packets are treated equally: when there is an excess of traffic resources and bandwidth. This technique often results in
it is “tail dropped” by a switch or access platform, an effect packets being dropped in the core of the network, where
especially disruptive to TCP traffic, while also causing retransmission and network overhead impact large amounts of
dropouts in VoIP and video streams. traffic. In general, if a packet is dropped before it reaches its
destination, all of the resources it has consumed in transit are
Increasingly this “dumb-pipe” approach is impacting
wasted.
businesses and cellular operators who are migrating towards
packet-based connectivity and applications. Consider the
QoS Management at the Service Demarcation Point
diverse demands of three common classes of applications
To effectively enable carrier-grade services over packet-based
sharing an access link – VoIP, IP-video and Internet browsing
infrastructures, service providers need to implement a QoS
– VoIP is a relatively low bandwidth application that is highly
management (traffic conditioning) scheme that ensures that
sensitive to both delay and jitter; video is less sensitive to
critical, real-time applications are allocated proper priority from
delay but is highly sensitive to jitter with high bandwidth
end-to-end. As all network resources are impacted by poor
demands; general Internet browsing is often bursty and may
traffic management, it is ideal to apply traffic conditioning
require large amounts of bandwidth, but is not often
techniques at the service edge, effectively reducing congestion
considered QoS-sensitive. Giving each of these applications
before it can enter the provider’s network.
equal access to limited bandwidth will impact the QoS of both
the voice and video applications, in the worst case resulting in
periodic service disruption as Internet usage varies over time. One natural way to achieve this to apply QoS management at
each service demarcation point, often defined by a Network
For providers offering Service Level Agreements outlining Interface Device (NID) located at customer premises, cell sites
QoS requirements, a best-effort access connection is not only and hand-off points between interconnecting networks. By
sub-optimal, it is also expensive: SLA penalties can quickly conditioning traffic at these network boundaries, service
wash out the extra revenue SLA-based services afford, and providers assume control over traffic entering and leaving the
customer churn rates will reflect poor service QoS and network, enabling the best possible QoS and assuring
availability. revenue-driving SLAs.

Hub

MSC
Cell
Wireless
Site Core /
Backhaul
Transport
Network CO
Customer
Metro /
LAN
Access Aggregation Hand-Off Partner
Customer Node Point Carrier
LAN

Typical NID Demarcation / H-QoS Application Points:

© 2008 Accedian Networks, Inc. ACCEDIAN.com


White Paper
ACCEDIAN Establishing Real-Time Hierarchical
NETWORKS QoS with Zero-Latency Traffic Shaping
PACKET PERFORMANCE ASSURANCE TM

QoS Management Techniques Hierarchical QoS and Traffic Shaping


A common, simple form of QoS management is bandwidth Hierarchical QoS (H-QoS) schemes are designed to provide
policing, also known as rate limiting. Affectionately referred to highly efficient use of access and network bandwidth by
by operations staff as “crushing the edge”, bandwidth policing implementing traffic shaping techniques in addition to simple
simply drops any excess packets that exceed the policy rate limiting. Unlike bandwidth policing that simply drops
defined for a particular flow. A typical bandwidth policy excess traffic, shaping is designed expressly to avoid packet
reflecting the coexistence of VoIP, IP-video and Internet traffic loss by selectively buffering traffic, effectively smoothing out
over a 10 Mbps connection is shown in the table below: bursty traffic to “right-size” it to the bandwidth available.

When used for H-QoS, shaping schemes rely on a hierarchy


of application priorities, which determine the degree of
buffering or policing applied to a particular traffic flow. Also
know as application acceleration, shaping can ensure that
the most QoS sensitive traffic always receives the bandwidth
required by “stealing” it from lower-priority flows. Benefits of
H-QoS to service providers include more efficient use of
network resources, the ability to adapt their network to reflect
the needs of their customers, fewer QoS issues, and a
Triple-Play Bandwidth Policy Example
greater capacity for over-provisioning.
This simple scheme ensures that VoIP and video maintain a
certain level of QoS for most practical scenarios, while Internet
traffic only consumes bandwidth not required by these QoS- Policer
sensitive applications. It should be noted that video is given a
Link Speed
small amount of EIR to accommodate a sudden increase in
Customer Provider
video traffic (e.g. when streaming starts) to improve channel LAN Network
change and service response time. However, there are
shortcomings to simple bandwidth regulation. If the demand for
Packet Loss -
a service increases, or the profile doesn’t accurately reflect affects all flows
service requirements, QoS will suffer. For example, if VoIP
traffic exceeds the policies set for it (due to more calls than Bandwidth Policing
expected, or new phones added without updating the policy),
Shaper
packets will simply be dropped, resulting in immediate call
quality issues. Link Speed
Customer Provider
LAN Network
Conversely, this policy does not allow Internet traffic to use
more bandwidth, even if VoIP and video data is not present.
In addition, this scenario requires that the aggregate bandwidth Limited, Prioritized Priority
Packet Loss
required for the worst-case scenario is always available – an Other
expensive form of QoS assurance, as applications are often
Traffic Shaping
bursty in nature, leaving a link underutilized (and overpriced) a
good part of the time.

© 2008 Accedian Networks, Inc. ACCEDIAN.com


White Paper
ACCEDIAN Establishing Real-Time Hierarchical
NETWORKS QoS with Zero-Latency Traffic Shaping
PACKET PERFORMANCE ASSURANCE TM

Traffic Shaping Limitations


Key Causes of Shaping Delay
As useful as shaping sounds in theory, service providers have
been justifiably reluctant to apply this technique to critical
 Packet Characterization Delay
traffic, as most shaping methods add significant delay to QoS-
sensitive traffic. In most shaping implementations two key  Store-and-forward, network processor based
software algorithms cannot process packets
aspects cause this unwanted delay:
without some inspection delay

(1) all shaped traffic (including real-time traffic) is queued in a  Silicon-based, hardware packet processing
architectures characterize packets at wire speed,
store-and-forward buffer for packet characterization and
eliminating characterization delay.
processing, and;

(2) common shaping algorithms, such as the IETF-specified  Shaping Algorithm Efficiency
Random Early Detection (RED) method, rely only (and  Standard algorithms, including RED, require large
directly) on queue length to estimate traffic congestion. buffers to adapt properly to bursty traffic, directly
affecting queuing latency.
Compounding the problem, these two key factors are
 The more efficient BLUE algorithm estimates
interrelated: the RED algorithm requires the use of large congestion using adaptive techniques that consider
buffers to work effectively with bursty traffic, increasing the both queue usage and link idle time, greatly
duration packets are delayed in priority queues 1. reducing buffer size, delay and packet loss.

Zero-Latency & Delay-Optimized Shaping Techniques


To shape traffic, packets first must be inspected and
characterized (the first source of delay), then priority queued
Customer Zero-Latency
for transmission (an additional delay dependent on buffer LAN Fast-Thru™
size). With these constraints, it is commonly assumed that
traffic shaping always introduces delay (and to a lesser
degree, jitter) to real-time applications, necessarily imparting Custom
an impact on QoS.
Silicon
(FPGA / ASIC)

Recent advances in packet-processing techniques and more


efficient shaping algorithms have been able significantly
Provider Priority
reduce these effects, overcoming the limitations of traditional Real Time Network Other
shaping methods and allowing this technology to become a “On the Wire”
key part of H-QoS schemes for even the most delay-sensitive Packet
applications. The best possible shaping implementations Processing
today add inconsequential amounts of delay and jitter to high-
priority traffic, while reducing the latency (and the size of Silicon-Based Packet Processing
buffers) required to queue less critical flows. & Zero-Latency Shaping

1 Feng, Kandlur, et al. Blue: A New Class of Active Queue Management Algorithms, 2000

© 2008 Accedian Networks, Inc. ACCEDIAN.com


White Paper
ACCEDIAN Establishing Real-Time Hierarchical
NETWORKS QoS with Zero-Latency Traffic Shaping
PACKET PERFORMANCE ASSURANCE TM

Packet inspection delays can be virtually eliminated by silicon- Instead of relying directly on queue length as an indicator of
based packet processing architectures that, unlike network congestion, BLUE uses an adaptive algorithm that considers
processor-based designs, do not require store-and-forward both queue usage and link idle events to manage traffic
buffering to characterize packets. Dedicated, silicon-based queues, resulting in both significantly lower packet loss as well
processing is performed at full wire-speed, introducing < 5 μs as smaller buffer-size requirements. Smaller buffers result
of delay, up to 100 times faster than alternative designs and directly in reduced shaping latency, while less packet loss
1,000 times faster than the 3-5 ms end-to-end, one-way delay decreases the utilization of network resources. This overall
limit often specified in real-time application SLAs. Since improvement in shaping performance is due to the BLUE
algorithm’s proven ability to effectively adapt to sudden
hardware-based processing is done “on the wire”, the highest-
changes in bandwidth demand.
priority flows can continue in transit without any buffering or
shaping delays. This “fast-through” technique is also known Implementing Zero-Latency Traffic Shaping & H-QoS
as a real-time queue, akin to a dedicated carpool or bypass at the Service Demarc
lane on a highway, effectively resulting in zero-latency shaping
With these new advances in traffic shaping technology, service
impact for the most sensitive traffic.
providers can now implement high-performance H-QoS for
even the most demanding applications. By using network
Even with the availability of a real-time queue, lower-priority interface devices (NIDs) to perform these traffic conditioning
traffic still needs to be buffered and processed to shape the functions at the demarcation point, providers can most
remaining traffic. To reduce queue-related delays, a new effectively take control their services from edge-to-edge, even
shaping algorithm known as BLUE has been developed that over multi-vendor, multi-carrier and multi-technology
offers significant performance improvements over the converged networks, without reconfiguring, upgrading or
commonly implemented RED algorithm 1. replacing existing access platforms.

MetroNID™ Demarc Units with Traffic Shaping

Accedian Networks’ MetroNID™ packet assurance


demarcation units combine silicon-based packet processing,
the BLUE shaping algorithm and high-performance bandwidth
policing in a compact, cost-efficient unit, allowing service
providers to effectively apply H-QoS at customer locations, cell
sites, and hand-off points throughout their network.
METER

Demarc & In-Service Assure


As the devices also provide a wide range of service creation Loopback RFC-2544 SLAs
and service assurance capabilities, providers using MetroNID
units can quickly establish and assure large-scale SLA-based
Ethernet & IP services over any network topology, while
optimizing real-time application performance. Test Set EVC Shaping,
Support Creation Policing

1 Feng, Kandlur, et al. Blue: A New Class of Active Queue Management Algorithms, 2000

Accedian Networks Accedian Networks’ Packet Performance Assurance solutions enable carrier-grade, packet-based
4878 Levy St, suite 202 wireless backhaul, business services & multi-carrier applications over wireless & wireline networks.
St-Laurent, QC, Canada, H4R 2P1
+1 (514) 331-6181, 1 (866) 685-8181 © 2008 Accedian Networks, Inc.
Version: June 08

Accedian, EtherNID, MetroNID, SLA-meter, Performance Assurance Agent, PAA, Multi-SLA, Ether-PRO, EtherSHELF, Ethernet Service Assurance Platform (ESAP), EVC-Builder,
“Packet Performance Assurance”, “Delivering Service Performance with Assurance” and “Map-out the Health of your Network” are trademarks of Accedian Networks Inc.
ACCEDIAN.com

You might also like