IP telephony has proven its value
in large corporations, and as wireless LANs become more commonplace, companies
are now investigating the possibilities of making IP phones mobile through voice
over WLAN (VoWLAN) technology. WLAN industry vendors are stepping up to the
plate with many different products. But VoWLAN presents unique new challenges
for those deploying WLAN systems, and the products that traditional WLAN vendors
have offered for hotspot coverage in data networks may not work well -- or at
all -- when asked to manage voice traffic as well.
Predictability is the key to good
voice communications. Voice services are very sensitive to digital transmission
errors. When voice is transmitted, a specific compression/decompression (CODEC)
scheme is used to compress the outgoing traffic, which is reconstructed on the
opposite end of the transmission by a matching CODEC. During transmissions,
however, lost packets, random network delays, and retransmissions increase
"jitter" (or variation in timing, or time of arrival, of received
signal) and signal loss, causing clicks, silent periods, and poor voice quality.
When more advanced compression techniques are employed, the problems become even
more pronounced because these techniques further delay transmission of the
signal.
VoIP systems use buffers to smooth out minor variances in packet transmission
rates, and these buffers work fairly well in wired networks. But WLANs feature
wider variances in transmission speeds, and delivering packets predictably is
where traditional WLANs fall short for mobile VoIP in an enterprise.
VoWLAN Challenges
WLANs today face two significant limitations that make them unsuitable for use
as a voice transmission network across an enterprise: they can't prioritize
and reliably deliver data packets both to and from the client, and they cannot
maintain predictable performance between the client and an access point (AP) as
the client roams from one AP's coverage area to another's.
Cellular equipment manufacturers have designed their base station technology to
solve these problems in full. In a cellular network, each base station
coordinates with the others to optimize the connection with each client as those
clients roam. These base stations have full control over the air interface with
each client, with the ability to control the ordering of packets both
transmitted and received to adjust precisely for the quality of the connection
with each caller. In most of today's wireless LANs, however, the
consumer-grade APs being used as base stations can't coordinate with other
APs, and they can't deliver such precision control over air traffic.
Almost universally, traditional Wi-Fi vendors have chosen to get to market
quickly by using consumer-grade AP designs. These vendors layer some software
intelligence on top of off-the-shelf APs to facilitate security, configuration,
and authentication, but the chipsets on which the APs are based have static
settings for channel access, transmission timing, and packet queuing functions.
The off-the-shelf chipsets in these designs serve exactly the purpose they were
created for: to make the AP system designer's life easy. The software need
only push packets to the MAC through a queue, and that the chip takes care of
the rest. This off-the-shelf AP technology works well for consumer and small
office environments, where only one or two APs are needed and only a handful of
clients are active. With appropriate management software, these same APs can
even work in multi-AP enterprise settings for data applications, albeit with
reduced performance. But the inability to control the quality of the connection
with each client is a critical impediment to providing suitable voice services
over a WLAN built on such AP technology.
Why Commodity
Access Points Can't Deliver Quality Voice
Basically,
off-the-shelf Media Access Control (MAC) chipsets take a stream--or multiple
streams--of packets, and use weighted preferences to determine which stream
should be transmitted first. These chipsets are responsible for loading the
packets in from memory at some point before transmission. Once the packets are
inside the chipset, the hardware then runs through the 802.11 contention
algorithms, varying some of the backoff numbers in accordance with the priority
of each stream (this is 802.11e's mechanism). All critical timing and packet
ordering issues are handled locally, within these off-the-shelf components. From
there, the packet is transmitted to the wireless network.
Building on this
common chipset base, WLAN vendors differentiate themselves with overlaid
software intelligence. Some deal largely with mobility and configuration issues.
The most advanced vendors provide automated control of basic RF configuration
settings (channel, power, etc.), automating much of network management and
security, and providing packet-reordering and classification services before
presenting the packet stream to the off-the-shelf chipsets.
The problem with this architecture is that the middleman, the off-the-shelf
chipset, holds the system designer at bay. APs based on these chipsets have
access to broad strokes of wireless information such as signal strength, station
identification, link status, and other asynchronous information, allowing the
WLAN controller to set the aggregate transmit powers for the cells, for example,
and to herd clients onto different APs on the basis of some optimization
criteria. But the WLAN controller has no control over the details of the air
transmission (the latency, jitter, and error rate) because this information
isn't being presented to the AP's software and therefore isn't being
relayed to the centralized control unit. And it is precisely these details--the
latency, jitter, and error rate--that must be coordinated to ensure predictable
quality of service for voice.
Without having detailed, fine-grained control over the air, APs using these
chipsets habitually deny voice clients their chance to transmit. Data traffic,
being much faster and more aggressive, overwhelms the slow-and-steady voice
calls because the collision-avoidance algorithm in WLANs (CMSA/CA
-- see sidebar) responds to the
most aggressive incoming traffic. On the downstream side, the opportunistic data
flows jam the MAC chipset, and quickly overwhelm its ability to guarantee any
particular flow rate for voice.
Just as off-the-shelf APs make predictable packet delivery impossible, they also
prevent reliable handoffs when the user roams from one AP's coverage area to
another's. The off-the-shelf chipset in each AP must act as an independent
entity in terms of managing the air connection, so in this architecture the
client makes the decision of when and where to hand off. A client that switches
from one AP to another is forced to perform a detailed, multi-packet handshake
each time it re-associates. This handshake has many components, and if any part
is lost due to congestion or interference, the entire handoff may collapse.
Should the handoff collapse, a long, painstaking process ensues, where the
client first must reenter the network by choosing an access point, and then must
re-present all of its credentials--especially if VPNs are used to secure the
network--causing multiple seconds of handoff delay in the worst cases.
A Better
Solution Starts With Better Hardware
Fortunately, the
802.11 standard enables innovation and encourages methods by which vendors can
design enterprise-grade access points that do provide precision control over the
quality of client connections, and which thereby solve the problems of voice
quality.
Precise control over the timing, transmission, and packet-by-packet scheduling
provides for traffic management over the air not only for each AP, but also for
coordination across APs. With these parameters exposed to the system designer,
it's possible to create a WLAN system that maintains a global knowledge of the
RF environment. Such a WLAN system can coordinate transmissions to provide
crystal clear voice communications for every client without fear of interference
from other local transmitters such as cordless phones. In addition, the same
global knowledge of the RF environment allows system designers to eliminate
interference from neighboring APs on the same RF channel.
By asserting
control over the MAC itself, access points can cooperate--without knowledge or
involvement of the client--to seamlessly transfer ownership of a connection.
When a dynamically controlled MAC is used, the problem of maintaining voice
quality during handoff becomes a simple matter of having the central controller
choose which AP is best. In other words, the wireless infrastructure decides
which AP is best for the client at any given time, rather than the client
deciding which AP is best.
Because the coordinated APs and wireless controller are maintaining awareness
and control of the whole environment and the quality of service to each client,
this approach brings order and predictable service quality to an otherwise
random communication medium. As such, this coordinated AP approach resolves the
voice quality and Wi-Fi handoff problems in a similar fashion to the proven
method used in cellular networks.
As WLANs are tasked to carry more traffic for more types of applications, both
delay-sensitive and otherwise, it will be essential for WLAN vendors to use
hardware architectures that provide precise intelligence about and control over
the quality of each AP-to-client connection. While off-the-shelf AP chipsets may
have been good enough for hotspot coverage to date, they won't work for the IP
voice applications or the pervasive WLAN coverage companies envision in the
future. More intelligent WLANs must begin with more intelligent WLAN access
point hardware.
About
the Authors
|
|
Kamal Anand is VP of
Marketing at Meru Networks. Previously, he served as VP of Asia Pacific
Sales and Global Business Development with Atoga Systems, which
developed systems for carrier customers. Before that, he was VP of
Marketing and Business Development and a member of the founding team at
NetContinuum, an emerging provider of security products for Enterprises.
Kamal has also served as VP of Worldwide Field Marketing at Marconi
Communications/FORE Systems, which he joined through the acquisition of
Berkeley Networks.
|
|
|
Joel Vincent is Product
Marketing Director at Meru Networks. Previously, he played instrumental
roles in the founding of both the Lucent Netcare network management
software product line and the NETGEAR consumer networking products. He
also helped CopperCom launch the first commercially deployed voice over
DSL system and developed Conxion Corporation's managed hosting service
suite. Joel has a BSEE degree from MIT as well as Sloan School of
Management experience.
|
About Meru
Networks
|
|
Meru
Networks designs and develops
standards-compliant 802.11 Wireless LAN Systems for large-scale data,
voice and real-time applications. Meru's products include coordinated
Access Points (APs) and Controllers that manage multiple APs. The Meru
solution, deployed in Fortune 500 companies, universities, and
healthcare organizations, provides over-the-air Quality of Service,
predictable performance, and roaming with seamless handoffs. Meru's
Wireless LAN System greatly simplifies RF planning associated with
large-scale WLAN deployment and provides the industry's most
comprehensive WLAN security from location to application, with
continuous monitoring. Meru also provides a comprehensive set of network
management tools to minimize ongoing operational costs.
|
