May 2004
Is Your Network Ready For Convergence?
BY
ARVIND AHUJA
Is your data network ready to carry packetized voice traffic on top of
all your current applications? For most enterprise network managers, the
likeliest answer would be: I don�t know. That�s the make-or-break question
for many network executives as they consider the promise of integrated
voice/data.
The biggest challenge in transitioning from traditional circuit-switched
voice and video systems to the new, more economical voice and video over IP
packet-switched technologies is obtaining adequate quality of service (QoS)
over the network. Quality of service is the capability built into the
network to guarantee that information traverses the network in a timely
manner. Most existing data networks were designed for bursty applications
that are delay-insensitive, meaning that if a data packet arrives within a
reasonable amount of time, both the application and the user are satisfied.
Voice and video data, on the other hand, are very sensitive to delay; if
a packet arrives more than approximately 170 milliseconds (ms) after it is
transmitted, the packet is worthless as it will arrive too late to be used
in the conversation or video image. Consequently, networks carrying IP voice
must be designed and configured properly to ensure that real-time packets
traverse the network efficiently.
The challenge of obtaining adequate quality of service is exacerbated
when a data packet must traverse the WAN. Typical local-area networks (LANs)
run at 10 Mbps, 100 Mbps, 1000 Mbps, and higher. However, because bandwidth
over the WAN is significantly more expensive than over the LAN, many
wide-area networks operate at T1 speeds (1.45 Mbps) and slower, creating a
huge bottleneck at the LAN/WAN interface.
For normal data packets like e-mail, Web browsing, client/server
programs, and a host of other applications, this LAN/WAN bottleneck is a
nuisance, but not an application killer. However, when voice and video
packets must compete with regular data packets for transmission over a
bandwidth-constrained WAN, voice and video applications are often rendered
useless.
IMPORTANT QoS PARAMETERS
For IP networks supporting voice, video, and data applications, the network
quality of service is evaluated by measuring four key parameters: bandwidth,
end-to-end delay, jitter, and packet loss.
- Bandwidth: The average number of bits per second that can travel
successfully through the network.
- End-to-end delay: The average time it takes for a packet to traverse
the network from a sending device to a receiving device.
- Jitter: The variation in end-to-end delay of sequentially transmitted
packets.
- Packet loss: The percent of transmitted packets that never reach the
intended destination.
Target values for delay, jitter, and packet loss are <170 ms, <50 ms, and
<1% respectively. Organizations wishing to maintain management control of
their networks when adding new applications, including voice and video over
IP, usually implement control procedures using a number of available
technologies.
IMPLEMENTING CONTROL PROCEDURES
There are a number of control-based measures to provide adequate quality of
service.
IP Precedence and DiffServ
Both of these traffic-marking schemes modify and mark certain bits in the
data packet header. Upon arrival at an IP precedence or DiffServ-enabled
router or switch, packets with the header bits set appropriately are given
priority queuing and transmission.
In a network controlled by packet marking, voice packets would be given the
highest priority since they are very sensitive to delay and jitter, even
though voice is not particularly bandwidth-intensive.
Queuing
Queuing occurs in routers and switches. Different queues or buffers are
established for the different packet-marking schemes. One of the queues, for
example, might be established for delay- and drop sensitive information like
voice and video data. Voice and video packets marked with certain IP
precedence or DiffServ values will be placed in these high-priority queues.
TRAFFIC SHAPING
Queuing-based solutions have a number of drawbacks. Of these, one of the
most significant is the lack of any feedback mechanism for determining how
applications are competing for bandwidth. Consequently, data traffic for
applications on networks with queuing mechanisms in place cyclically ramp up
and back off transmission rates based upon packets being discarded. This
causes chunks of data that accumulate at the LAN/WAN interface where speed
conversion occurs.
One way to eliminate these chunks of data is by using a special technology
called TCP Rate Control. TCP Rate Control paces or smoothes network data
flows by detecting a remote user�s access speed, factoring in network
latency, and correlating this data with other rate and priority policies
applied to various applications. Rather than queuing data in a switch or
router and metering it out at the appropriate rate, TCP Rate Control induces
the sending applications to slow down or speed up, thus sending data
just-in-time. By shaping application traffic into optimally sized and timed
packets, TCP Rate Control can improve network efficiency, increase
throughput, and deliver more consistent, predictable, and prompt response
times.
Most voice and video applications use UDP rather than TCP for
transmitting real-time communications data. Unlike TCP, UDP sends data to a
recipient without establishing a connection, and UDP does not attempt to
verify that the data arrived intact. Therefore, UDP is referred to as an
unreliable, connectionless protocol. The services that UDP provides are
minimal � port number multiplexing and an optional checksum error-checking
process � so UDP requires less processing time, and lower bandwidth overhead
than TCP. This allows UDP packets to traverse the network more rapidly,
which is a desirable characteristic for voice and video applications.
However, because UDP doesn�t manage the end-to-end connection, it does
not get feedback regarding transmission conditions; consequently,
applications transmitting UDP packets cannot prevent or adapt to congestion.
Therefore, UDP can end up contributing significantly to an overabundance of
traffic impacting all traffic on the network. This may cause
latency-sensitive flows, such as voice and video over IP, to be so delayed
as to be useless. In these instances the voice or video application may
still continue to transmit data, oblivious to the fact it is contributing to
the delay problem.
PARTITIONING
A process called �partitioning� can control the rate of UDP transmissions
over the WAN.
Partitioning is a special case of rate control in which specific amounts
of bandwidth are set aside for the most important classes of traffic.
Partitioning can also be overlaid on top of TCP rate control for TCP based
applications. A packet-shaping appliance that examines all packets
traversing the network administers partitioning. By identifying a particular
application as a member of a particular partition class, the Shaping
Appliance is able to control how much bandwidth each application or class of
applications uses, and it can ensure that a particular partition always gets
sufficient bandwidth.
When the bandwidth within a particular partition is not fully utilized,
the excess bandwidth can be reallocated to partitions serving other
important applications. Administrators can also specify UDP flow maximum
mechanisms so that one large flow (e.g., video conferencing or streaming
video) does not consume all bandwidth on the network. In environments where
multiple video devices are deployed, organizations will want to couple
partitioning with a SIP proxy�s or H.323 gatekeeper�s call bandwidth
controls to manage how many simultaneous video calls can be placed.
CONCLUSION
Transitioning to voice and video over IP requires rock-solid quality of
service over the wide-area network link. One of the major IP packet
congestion points is at the LAN/WAN interface, where bandwidth availability
can often decrease by two or more orders of magnitude. Provisioning
additional WAN bandwidth may provide temporary relief, but it may be an
expensive short-term solution because additional bandwidth is often consumed
by more aggressive non-business applications like Web surfing and
peer-to-peer file sharing.
Arvind Ahuja is senior product manager for Packetwise software at
Packeteer, Inc., a leading provider of application traffic management
solutions designed to enable businesses to gain visibility and control of
networked applications, extend network resources and align application
performance with business priorities. For more information, visit the
company online at www.packeteer.com.
If you are interested in purchasing reprints of this article (in either
print or HTML format), please visit Reprint Management Services online at
www.reprintbuyer.com or contact a representative via e-mail at
[email protected]
or by phone at 800-290-5460.
[ Return
To The May 2004 Table Of Contents ]
|