After committing to deploy Telepresence, one of the first decisions an enterprise needs to make is whether they will carry the video traffic on their data network (converged) or whether they will build a separate WAN network exclusive to the video streams (overlay). Each choice has pros and cons so there is no right answer, they have to be evaluated for their relative merit within the context of each enterprise. Let's take a look at the value provided by each.Overlay Network: To create an overlay network, an enterprise contracts with a WAN service provider to create an MPLS network connecting those enterprise locations where telepresence and/or video conferencing will be used.
One of the biggest advantages of this approach is its simplicity: Create a network that has enough bandwidth on each access link to support the deployed telepresence or video conferencing. Implement two levels of QoS to keep management traffic away from the real-time traffic. Done. I often recommend this approach to enterprise IT teams who have an edict from on high to get this telepresence stuff working quickly.
Note that even though the pipe will be about 98% filled with real-time traffic (video and audio) it still needs at least a two level QoS because we don't want the management traffic to interfere with the video. Web-based management in particular can create sudden bursts of traffic which will cause queues to fill up momentarily, resulting in jitter or packet loss.
Lastly, with an overlay network it is possible to use a separate vendor for video than for data traffic. If the existing data WAN service provider is not ready for the demands of high bandwidth real-time traffic, create the overlay network with a vendor who is. Smaller boutique vendors who specialize in video do a great job with these overlays. And because you don't need the video to go everywhere, their lack of a complete global footprint may not be a problem.
The key disadvantage to the overlay approach is that the bandwidth is dedicated to video. When the video or telepresence systems are quiet, the network is quiet, and cannot be used for another purpose. And of course dealing with two network connections at each site, and possibly two vendors increases the complexity of managing the network and thus the cost of the complete solution. I often advise enterprises to start with an overlay network to get the deployment in place quickly and with high quality, as mentioned before. But then over time the team can evaluate how to migrate this traffic back to a converged network solution.
Converged Network: In the converged network approach we assume that video will become another application class and will ride on the same WAN as the rest of the network traffic of the enterprise. QoS will be deployed to ensure that the video traffic is delivered with appropriately low packet loss and jitter to ensure high quality results.
The primary advantage of a converged network solution is that there is only one network. This means fewer contracts, fewer connections, fewer service teams to deal with, fewer performance monitoring interfaces and one neck to choke. Keeping it simple on the contracting and vendor interface side brings a lot of advantages.
A second key advantage is that once deployed, the class of traffic supporting video can be tapped into anywhere in the organization. So unlike the overlay model, where we drop new lines into the six sites where telepresence will be deployed, with the converged model we have the video class available in all enterprise locations. As the use of video conferencing increases, incremental deployment of either telepresence suites or conventional video conferencing systems can be done easily.
The disadvantages of a converged network are primarily around the complexity of properly implementing a good Quality of Service strategy and determining appropriate bandwidth allocations for each class.
The QoS strategy must take into account all the applications running on the converged network, and assign them to an appropriate QoS class. The leading edge WAN service providers are moving to a 6-class model to allow a specific class for video. Ask your vendor if and when this capability will be available. Also take a look at IETF RFC 4594 for a good overview of how to assign applications to classes. Applications need to be assigned to priority classes based on the application's needs, not based on their importance to the enterprise.
Bandwidth allocation is straightforward for voice and video as it can be directly calculated based on the number of concurrent calls needed in each link. Calculating the right bandwidth allocation for data applications is harder because of their bursty behavior and because performance can be affected by limiting those bursts. Getting a good handle on the tradeoff of bandwidth and perceived performance by the users is a tricky task. Here is where WAN accelerators can help, as I discussed in a recent posting. Measurement of application performance can be done using the Apdex methodology, as recently discussed here.
Compromise: It may be possible to get some of each solution by creating an overlay network using the existing WAN service provider. The overlay network may even be able to operate in the same physical access links if there is sufficient bandwidth available. This approach can provide the advantages of an overlay network without increasing the number of vendors responsible for supporting the corporate WAN needs.