Global Sources
EE Times-Asia
Stay in touch with EE Times Asia
EE Times-Asia > Networks

How to engineer VoIP on an enterprise WLAN

Posted: 17 Oct 2005 ?? ?Print Version ?Bookmark and Share

Keywords:voice traffic?

By Edward Lor
SiNett Corp.

Market research studies clearly indicate that voice over WLAN (VoWLAN)!a marriage of VoIP and 802.11-based WLAN!will be a very important application in enterprise networking.

With the expected high-volume usage of VoIP-enabled WLAN devices and VoWLAN handsets, the need for a next-generation networking infrastructure to support VoWLAN is clear. As the industry heads towards a single network solution, next-generation enterprise networks will see unified accessibility for both wireless and wired clients, as well as the convergence of voice and data applications.

One of the key issues in legacy VoIP is traffic engineering.

Because voice traffic is delay sensitive, it requires preferential treatment. In enterprise wired IP telephony, most legacy network infrastructures offer some types of quality of service (QoS) functionality to accommodate the timely delivery of voice traffic to its destination.

The edge equipment!mainly the current generation of intelligent switches!includes features such as packet filtering, class-of-services, egress priority scheduling, bandwidth guarantee, etc., to accommodate voice traffic. These issues are well understood in legacy wired IP telephony.

Beyond the wired backbone
In VoWLAN, all the issues in wired VoIP are applicable. However, the wired segment only represents the backbone of the network. The wireless segment and the wired/wireless interface represent new challenges to the traffic engineering problem because they become the intermediate zone between the network backbone and the actual clients.

There is a discrepancy in bandwidth capacity between the wired segment and the wireless segment. Because the wireless segment's bandwidth is considerably lower, it is essential for the wired network to control the flow of traffic into the wireless network.

Furthermore, since a wireless client is mobile, its connecting point to the network may change from one access point (AP) to another. An intelligent switch supporting VoWLAN must offer the flexibility to efficiently move traffic from one connecting interface to another.

This article explores the traffic engineering issues in legacy wired IP telephony and the new wireless IP telephony. It will also focus on the requirements for enhanced traffic engineering in enterprise VoWLAN.

Multi-layer intelligent switching today
First-generation network switches have no intelligence. They cannot discern the type of traffic being transmitted (e.g., FTP versus HTTP). This is problematic because some packets require higher transmission priorities than others.

More specifically, VoIP traffic must reach the IP phone within a very small, very specific time window. If the packets are delayed due to network congestion, there will be an interruption in the conversation. As these first generation switches have no ability to distinguish voice traffic from regular data traffic in a congested network, the lack of traffic preferential treatment, or QoS, will negatively impact the IP telephony operation.

In the new generation of switching technology, multi-layer intelligent switches contain more intelligence in the switching hardware and software and are thus able to identify the type of traffic in order to place the packets into appropriate priority queues.

Multi-layer intelligent switches are aware of the content of every traffic flow inside the switch and can assign higher priority accordingly. In order for these switches to be aware of the content of each and every packet, they must be capable of a few basic traffic engineering functions-ingress filtering and egress scheduling.

Ingress filtering and rate policing
Ingress filtering occurs when the packets first come into the switch. It refers to the hardware-based identification of the type of traffic that is being transmitted. Every type of traffic has some form of identification in the packet&mash;the IP packet header and the TCP (transmission control protocol) or UDP (user datagram protocol) header.

The IP packet header identifies the sender (source IP address) and the receiver (destination IP address) of the packet. The TCP/UDP header indicates which application in the source is sending the packet and which application in the destination is supposed to receive the packet.

More specifically with ingress filtering, the intelligent switch tries to identify the content of the traffic and determine the type of application that is being sent by looking at the port numbers in the TCP/UDP header.

In relation to priority scheduling, ingress filtering allows the switch to determine if the packet is a voice application or a data application and assign priority accordingly (i.e., higher priority for voice packets and lower priority for data packets).

Priority is not only given as packets enter the switch at the ingress. After the switch has determined the content of the packet, the next step that occurs in intelligent switching is called egress scheduling. That is, high priority packets are placed in high class of service (COS) queues and lower priority packets are placed in lower COS queues as they exit the switch. Commonly used scheduling algorithms include strict priority scheduling (higher COS queues always get served ahead of lower COS queues), weighted round robin (certain number of packets from each COS queue get served, with higher COS queues getting more packets served), etc.

Rate policing, which limits the rate of traffic that comes into the switch, is another aspect of basic traffic engineering. Certain applications may send an excess amount of traffic into the network and congest it.

In multi-layer intelligent switching, the smart switch can avoid such applications jamming the network by limiting the rate at which packets can come into the switch. For example, a PC may be sending traffic to the network at 50 Mbps. However, the switch may not be able to handle that much traffic from one PC as it is also responsible for handling traffic from other PCs, APs, wireless clients, etc.

Rate policing allows the switch to limit the rate of traffic entering the switch to the rate determined in the service level agreement. Further, the network administrator can create various policies to determine what will happen with any extra bandwidth.

For instance, a policy can be created to allow the excess traffic to pass through the network if the network is not congested at the time of transmission. However, as soon as the network is congested, the packets may then be dropped. Conversely, a policy may be created to drop the excess traffic at all times.

Next generation: Unified access
Ingress filtering, egress scheduling and rate policing are all requirements for effective traffic engineering in current generation intelligent switching. Though they work well in wired VoIP application, VoWLAN presents new technical challenges.

Because voice traffic is latency sensitive, the timely delivery of voice traffic to its destination is still of utmost importance. Thus, in addition to having ingress filtering, egress scheduling and rate policing capabilities, next generation intelligent switching silicon must have additional features to enhance the VoWLAN experience.

In this illustration, the voice stream from the wired IP phone, Clienta, traverses through the unified access (UA) switch through a high priority queue. Even though the voice packets are sparse, due to a much lower bitrate, as compared to data packets, the high priority of these packets give them enough opportunities to be switched out. Such preferential treatment ensures that the voice traffic will reach the WLAN handset (Clientb) in a timely manner.

Egress scheduling in legacy multi-layer switching is referred to as static queue scheduling. That is, multiple COS queues are permanently associated with each switch port!it is static. However, static binding of COS queues to ports has its drawbacks in VoWLAN applications.

Centralized queue pool and dynamic queue binding
In a WLAN environment, each AP is plugged into a physical port on the switch. If the wireless client moves from one AP to another, voice packets waiting in a queue that is scheduled to be delivered to the first AP will not reach its destination handset. This will result in lost packets and degradation of voice quality.

In the VoWLAN context, the UA switch must support flexible or dynamic bindings between flow queues and egress ports, such that the packets already queued up will exit at the appropriate egress port. This feature is very crucial for the infrastructure to provide lossless mobility and timely delivery of voice applications.

Specifically, the binding scheme of CoS queues to egress ports must be revised for mobile-aware switching. In this solution, there is a centralized pool of egress CoS queues in the switch. The CoS queues are allocated on demand from the centralized pool as soon as there is a flow set up for ingress filtering of the voice stream.

The CoS queue is bound logically only to an egress port that is linked to the IP phone. For wired IP telephony, this binding will be unchanged for the duration of the call. However, for VoWLAN, this binding is subject to change as soon as the mobile client moves to another AP.

In this illustration, the queue CoSQx is destined for the flow of downstream packets to the wireless handset. Packets are accumulated in this CoS queue, which is initially bound to egress group Grpp. Each egress group, in essence representing an aggregation of all the traffic of a BSSID (Basic Service Set ID), is bound to an egress port. After the client hops from AP1 to AP2, thus changing its association to the switch from Portp to Portq, the binding of CoSQx to Grpp will simply be changed to Grpq.

Traffic shaping
Traffic shaping is another feature that is critical to the timely delivery of voice traffic. It is very similar to rate policing in that rate policing limits the bandwidth coming into the switch and traffic shaping limits the bandwidth going out of the switch.

The bandwidth of a wireless network depends on the 802.11 protocol used. If the protocol is 802.11b there is 11Mbps of theoretical bandwidth; whereas, if the protocol is 802.11g, there is 54Mbps of theoretical bandwidth.

If a 100mB port is pumping outbound traffic into the AP at 100Mbps, the AP will get bogged down very quickly due to the standard-based wireless bandwidth. One solution to this problem is referred to as traffic shaping. Traffic shaping limits the outbound traffic, for example to 11Mbps, making the traffic slow enough for the AP to handle and to distribute to the wireless clients.

Location awareness
Location awareness, also referred to as location tracking, is another essential function in VoWLAN traffic engineering. In order to support interoperability with cellular networks, the WLAN switch must be able to track the location of the dual mode cellular/WLAN phone at all times.

This knowledge is crucial to the handoff operation between the WLAN and the cellular network. With location identification and fluctuations of signal strength, a certain amount of pre-handoff setup can be done to ensure a smooth and speedy transition of the call session between the two different networks.

Further, location tracking is possible for different levels of granularity, from a high level (i.e., the AP coverage area a client is currently in), to a low level (i.e., the client's precise location within the coverage zone). High-level tracking is made trivial in most intelligent switches simply by filtering the ingress traffic based on AP origins.

However, low-level tracking is more complicated in that it requires switches that can interpret and process radio signal strength and support the typical triangulation method (signal strength measurement of the client from two or more bases). From a network topology perspective, it is the intelligent switch that is connected to all of the APs, coupled with the appropriate software that is able to perform these functions through location tracking.

With the ability to locate a device within an AP's coverage area, it is possible to further support advanced features such as pre-authentication and load balancing. Location tracking is also crucial in pinpointing a client device in case of a 911 emergency, particularly in a large enterprise premise.

Wired and wireless QoS standards
Another major traffic engineering issue is the difference between the wired QoS standards (802.1p, IP TOS, and DiffServ) and the wireless QoS standard (802.11e). Specifically, it is the responsibility of the wired/wireless interface equipment to map one standard to another.

For QoS in wired traffic, there are fields in the L2 and L3 frame headers that indicate the priority of the packet. For example, there is the 802.1p field, the type of service (TOS) field and the Differentiated Service Code-Point (DSCP) field.

Both the 802.1p and the TOS fields are three-bit fields, with a packet priority value of zero to seven (with zero being low priority packets and seven being high priority packets). DSCP, which has replaced TOS, is a six-bit field with values from zero to 63.

As previously stated, when a packet comes into the switch, the switch can perform ingress filtering to determine the priority of the packet. However, by reading the L2 and L3 frame headers, the intelligent switch can skip the lengthy process of going through each and every packet to look at the content, simply trust whatever priority the header reads and place the packets in the appropriate queues.

The previous generation of multi-layer switching can handle these priority fields and schedule according to whatever priority that comes with the packet. However, this is only limited to the wired world as wireless traffic presents a whole new set of standards. Specifically, 802.11e is the standard that deals with priorities in WLAN. A thin-AP/smart-switch model offers centralized management of QoS specifications for both wired and wireless traffic.

The switch can centrally configure uniform QoS policies for certain clients or applications, regardless of whether the traffic travels through the wire or over the air.

In addition, when the switch controls the QoS parameters, it can configure the 802.11e value in an 802.11 frame to be transmitted, according to a direct mapping between wired QoS and 802.11e. This feature relieves the AP of managing between wired QoS and wireless QoS, and saves the maintenance efforts required to keep them consistent across multiple APs.

In order to deploy VoWLAN, enterprise LAN infrastructures must be enhanced to accommodate various technical challenges associated with wireless networking and voice applications.

LAN networking equipment must be able to address several key issues in order to offer an effective integrated wireless/wired IP telephony solution. As previously stated, one key technical challenge is that of traffic engineering. The LAN infrastructure must be able to support advanced traffic engineering functions in order to accommodate the timely delivery of voice traffic to its destination.

SiNett Corp.'s OneEdge is an example of switching technology that can handle the next generation of advanced applications in enterprise communications. With features like ingress filtering, egress scheduling, rate policing, dynamic egress queuing, traffic shaping and signal strength management, OneEdge has the essential capabilities to efficiently address all the critical traffic engineering issues encountered in enterprise VoWLAN applications.

With the potential mass deployment of VoWLAN services in the enterprise, there is a need to examine the limitations of legacy switches to accommodate key issues such as traffic management.

About the author
Edward Lor
is a senior application engineer in SiNett Corp Before joining SiNett, he was the chief architect in the Wireless Communications Group of Hong Kong's Applied Science and Technology Research Institute. Lor has also held various architect, technical and project lead positions in Broadcom Corp. and Lucent Technologies. He also has extensive experience in IP telephony, VoIP security, wireless LAN and multi-layer networking. Lor received his Ph.D. in computer science from the University of California, Los Angeles.

Article Comments - How to engineer VoIP on an enterpris...
*? You can enter [0] more charecters.
*Verify code:


Visit Asia Webinars to learn about the latest in technology and get practical design tips.

Back to Top