Judging by the 1,600 users who registered for a recent audiocast I conducted with our test-tool partner, NetIQ Corp., network architects are desperate for hard data about how best to prepare for deployment of voice-over-IP networks.
The reality is that there are no definitive answers when it comes to deployment of voice over IP. There are only choices - lots of choices and lots of trade-offs you must make that have enormous implications for the quality of your voice services and the resources you consume delivering them.
Our primary message was the integration of voice over IP forces you to make trade-offs. There's a cost in the bandwidth expended for delivering the voice stream; a trade-off in the amount of latency you let affect voice quality vs. how efficiently you stuff voice samples into IP packets; and a trade-off in how secure you make your voice-over-IP transport vs. the resource burden that entails.
The traditional method of adopting a new technology such as voice over IP is to purchase voice-over-IP gear and run a pilot; stabilize the technology and then deploy it in production mode. But NetIQ's John Walker says while this tried-and-true approach worked well in data nets, it can't simply be adopted for voice over IP. He says, and I concur, you need to first determine if your data network fundamentally can support voice over IP - that is, have acceptable congestion, latency and jitter characteristics - well before you purchase your first piece of voice-over-IP gear.
That means going down the checklist of your network infrastructure components and testing them to ensure they support voice by not imposing unreasonable delay, contributing to packet loss or causing jitter.
And you have to test the voice encoder/decoders (vocoder) you intend to use. These are the devices that packetize voice traffic for travel across the IP network. You have to be the judge for what level of vocoder compression is required to deliver the audio quality best suited for your applications. That compression choice, which ultimately determines the base bandwidth of a voice stream, correlates directly to the number of voice-over-IP calls you can support across a given link.
A vocoder that supports the G.711 specification (that transmits voice at an equivalent of 64K bit/sec) doesn't fit 24 voice-over-IP calls into a T-1 line. Assuming two voice samples in each packet and taking IP packet overhead into account, you can fit about 16 calls onto that fat pipe. You can fit more calls onto the link by using vocoders with greater compression ratios, but then you run into latency and packet-loss trade-offs again.
Even if you're just thinking about voice over IP now, the more you know upfront, the better prepared you'll be to address myriad of trade-offs you'll face when it's time to make the move to voice over IP.