In order for your wide area network (WAN) to successfully carry high-quality voice traffic, it must have the following characteristics:
- low latency
Voice over WAN requires careful consideration of codecs, WAN optimization and bandwidth provisioning. But simply engineering the optimal network isn't enough; the final piece of the puzzle is implementing tools (or services) to proactively manage VoIP quality and alert you to problems before you get calls from angry users.
To control latency and jitter, you must move packets from one end of the WAN to the other reliably and consistently. As a general guideline, one-way latency should not exceed 150 milliseconds -- from the time the caller speaks to the time the person on the receiving end hears the voice. Jitter, the variation in delay, should be less than 20 milliseconds. But even if you mitigate latency and jitter, bandwidth contention is still an issue -- especially in the WAN, where bandwidth is still more expensive than in the local area network (LAN).
More on voice over WAN
Voice application delivery gets boost with app session controller
When does latency become an issue to the flow of voice traffic?
Choose the proper voice codec for voice over WAN
The first step in optimizing voice over WAN environments doesn't even involve the WAN. Rather, it comes in selecting a voice codec that can provide acceptable quality while minimizing bandwidth utilization. In the LAN, where bandwidth is plentiful, most companies use G.711 uncompressed voice, which requires 64 Kbps per call. In the WAN, thanks to the need for additional headers, G.711 can chew up more than 80 Kbps per call. As a result, many engineers look to compressed codecs such as G.722 or G.729 to reduce bandwidth use while still providing acceptable voice quality to the end user.
Deciding which codec to use will require careful testing with actual users. The result of your user testing may lead you to lower voice quality due to the high support costs of G.711 voice codecs. Companies wishing to use high-definition (or wideband) codecs will also have to balance the need for higher bandwidth against the benefits they receive from using higher-quality audio. Alternatively, some vendors offer proprietary codecs (like Microsoft RTAudio) that have the ability to adapt bandwidth use to available network bandwidth.
One popular approach is a hybrid model using G.711 for internal calls that go over the LAN while transcoding to G.729 for calls that traverse the WAN. This approach is especially useful for large campus environments where a large percentage of calls stay on the LAN or metropolitan area network (MAN).
Choose the best voice over WAN transport
Once you've picked the ideal codec, the next step is to tackle the actual transport of voice over the WAN. A variety of tools are available to minimize latency and jitter. We categorize these tools under the umbrella of application delivery optimization (ADO), which includes tools, protocols and services designed to meet the individual performance of specific applications across the WAN. It's important to note that when talking about ADO, we're not just optimizing for one application; rather, we're optimizing the entire WAN to support a range of applications, including voice, video, storage, backup, transaction processing, Web traffic and bulk data transfer. Meeting the needs of VoIP requires balancing overall WAN capabilities with the needs of other applications.
For most companies, sending voice over the WAN requires using a WAN service that can prioritize voice traffic ahead of other less latency-sensitive packets. Multiprotocol Label Switching (MPLS) is often a prerequisite since all MPLS service providers enable prioritization of VoIP on MPLS by using Quality of Service (QoS) and/or Class of Service (CoS). For those using Ethernet or other transport mechanisms, most router vendors support prioritization techniques that prioritize one class of traffic over another.
For those relying on the Internet as the WAN transport, your mileage may vary. We've certainly seen examples of successful implementations of voice over the Internet, and companies like Vonage and magicJack have built their entire businesses on this model. But it is impossible to guarantee voice performance over the Internet. For occasional use, traveling workers or telework, voice over the Internet may provide acceptable quality. For conversations with customers, business partners, and/or suppliers -- or important internal calls -- few companies will trust Internet performance for VoIP.
WAN services offer a variety of techniques to optimize voice. For example, simple prioritization and queuing schemes ensure that voice gets preferential treatment. Call admission control limits the volume of calls to avoid exceeding WAN bandwidth. Other techniques, such as rate shaping and caching, can reduce the amount of non-latency sensitive traffic before it affects voice quality. Many companies also rely on on-premises WAN optimization solutions to implement their own granular controls over application performance.
Voice over WAN management
The last component of a successful voice optimization strategy is management. There is an old adage in IT that applies here: You can't monitor what you can't measure. This is especially true for voice. Your voice traffic monitoring and management options include self-bought/provisioned platforms that report on actual voice call quality or create synthetic transactions to check network performance. Carriers and third-party managed service providers (MSPs) are also capable of providing the same level of insight into overall voice performance or metrics associated with specific calls.
Finally, it's important for network architects to understand that successfully optimizing voice over WAN is a continual process. As VoIP traffic increases -- and as voice extends to desktops and mobile devices -- so do voice quality management challenges. To ensure success, plan now for a solution that is based on:
- picking the optimal codec;
- controlling latency and jitter across the WAN;
- providing sufficient VoIP bandwidth for call volumes; and
- managing ongoing voice performance.
This was first published in February 2012