Telepresence is probably the toughest application on the enterprise network today. It requires better packet loss statistics than VoIP, tighter jitter requirements, low latency and enormous bandwidth. This combination provides a design challenge that enterprise network teams have not previously faced. And because the investment to set up telepresence suites is high, and the users are likely to be executive staff, the expectations for quality are also high and the tolerance for failure is low.
In many cases, the IT team is at a real disadvantage because they don't know what they don't know about this application. The sale often takes place in the CEO's office and the mandate to deploy it comes down from there. The schedule is tight because the execs want it now and want to show the value of the expensive decision they have made. All these components combine to provide a recipe for potential disaster.
So what does the design process look like? The first decision is to determine if the telepresence traffic will be carried on an overlay network or a converged network. An overlay network is a new set of connections that parallel the current network. These links are dedicated to the telepresence application. This decision (overlay versus converged) is driven by the sophistication of QoS in the current network and often by the deployment schedule. It is much faster to get an overlay network running correctly than to get all the details of additional bandwidth and QoS working on the converged network. I often advise enterprises to start with this approach, and work their way back to a converged network when they are ready.
The next step is to determine bandwidth requirements. How much bandwidth will be required to support the proposed telepresence suites? Telepresence systems have a range of bandwidths over which they will operate, with varying degrees of quality as a result. Run some tests with the vendor to determine what quality level you really need. Then lay out the network map and determine which LAN and WAN links will need to support that bandwidth. Telepresence systems usually consume about 5 Mbps per screen for today's technology. So a 3-screen system requires 15 Mbps of continuous network bandwidth when in use.
Next steps are to ensure that QoS is properly deployed to guarantee high-quality transport for the telepresence video and audio streams. Interactive video conferencing is a real-time application, so it uses UDP and requires low loss, low latency and low jitter. Getting this wrong means displaying your network loss on 60" plasma screens to your top level executives.
I will take a deeper dive into these items and a few more over the next few posts. We need to figure out how multipoint conferences affect the network, how to connect to the concierge service, how to specify the right SLA (service level agreement) for the WAN service providers, how to instrument the network to ensure it continues to work well and how to bind multiple WAN service providers to get the right global footprint. Hang on as we dive into the details in the coming weeks.
