VoIP primer: How it works - and what the jargon means

Ever wondered how IP telephony actually works? What jitter means or how packet loss is another's gain? Elizabeth Biddlecombe explains all.

IP telephony is nothing short of a revolution. It works in a fundamentally different way to how telephone networks have carried our voice communications over the past 100 years.

Traditionally voice is sent as a continuous stream over an open circuit from caller to caller, in what is called 'circuit switching'. The longer the circuit, the higher the tariff. The longer the call, the greater the calling costs. Despite long silences, the call is rated for every second the circuit is open. Even the migration from analogue to digital circuits did little to change this model.

But of course, IP has changed that. Just as a web page can be broken up into 'packets', audio can be sampled with a digital signal processor (DSP), 'packetised' and sent out over an IP-based network as another data stream. The IP network may be a local area network, a company wide area network, a telco's core network or even the public internet. The packets that make up the audio stream may take different routes from node to node over the network in question but and are put back into order at the termination point to make up the audio that we recognise as conversation.

VoIP and packet switching makes for a more efficient use of network resources. But IP, of course, was not designed to provide a guaranteed quality of service (QoS) to the traffic it carries; it was originally designed as a way of diversely routing data to ensure redundancy in case network nodes were being jammed, crashed or blown up with a nuke.

An IP packet is formed when a router chops up a stream of digital information into manageable chunks. The router then places two addresses in the packet's header field - one for the destination, one for the originating device. After looking up the packet's destination in the number's address tables, the router forwards the packet to the next router in the chain.

Routers ping each other, to see which pathway is congested, completely blocked or relatively open. The packet is forwarded along the most suitable route and so on. When packets arrive at their destination, they can arrive in any order with significant delay. Transmission control protocol (TCP) arranges them in the right order.

Fine for databases and email and so on but not so good for voice. A caller's words must reach their destination in real time. Any 'latency' - the delay in delivering the voice stream from the speaker to the listener - degrades the quality of the sound as does 'jitter' and 'packet loss'. If packets are delayed by more than 250 milliseconds from end to end, a conversation can sound slightly unreal, as if the other party is on the moon. Jitter relates to this - it is the variation in the time between packets arriving. Some quick, some slow can cause a stilted conversation. Consequently, VoIP applications buffer packets to allow them to be processed more smoothly.

The problems of packet loss, delay and jitter are exacerbated when voice traffic intermingles with packets from other applications. This is particularly pronounced when the voice traffic traverses the internet. In these circumstances, packet loss is inevitable and no service provider can guarantee otherwise.

On private networks - LAN, WAN or in the carrier's core - application performance can be assured using two distinct quality of service parameters: DiffServ and MPLS. The former is a way of ensuring application performance in Ethernet networks, the latter in IP-based or IP-encapsulated wide area networks. With DiffServ, information about priority is places in a packets header; with MPLS and extra header field is attached. Both can tell routers what priority it has over other applications. With MPLS in particular, applications can be graded into a number of classes. Sometimes these are called Platinum, Gold, Silver and Bronze but some carriers are now offering many more classes of service (CoS) than these.

To ensure voice quality, it is critical to evaluate what other demands are being placed on the LAN and WAN that will prevent voice traffic getting the treatment it requires. Essentially, it needs to be prioritised over traffic from applications like email that can tolerate some delay. Voice is typically given the highest priority, in MPLS terms this is likely to be Platinum, while email may be Bronze.

What makes VoIP even more confusing is the varieties that it has spawned, each with its own technical criteria. Even in at a business-grade quality, there are three distinct types of VoIP.

LAN telephony or IPT is where a business runs its voice calls over its local area data network, originating from IP handsets or 'soft phones' (software running on a PC). Voice traffic will be converged with data traffic over the same internal wiring. A phone port will be identical to a data port. To control switching in this environment, either a media gateway or an IP-PBX is needed. When making calls external to the company, the IP-PBX converts them back into circuit-switched for delivery over the public telephone network. Some carriers now offer VoIP so that calls need not be converted into circuit-switched.

For companies with a number of locations and LAN telephony at each, again VoIP can be carried over the WAN without needing to be transformed into and back out of circuit-switched.

As some companies have only recently upgraded their PBXs, the business case for migrating to an IP-PBX is less clear. The LAN cabling and routers may need upgrading, new handsets will need to be purchased and a media gateway will need to be implemented.

Consequently, the first forays in VoIP by many large organisations is to use VoIP in the WAN, which is known a VoIP trunking, the second variety of the technology. This can save money on phone calls between locations since the voice runs over the existing wide area data links between buildings rather than being sent over a phone company's network. If remote workers or travelling employees dial into the data network using an IP-VPN, they can save money by making phone calls over the same virtual private network.

A company might opt to use LAN telephony if its existing PBX needs to be replaced or expanded or if a new building is being provisioned. Any company that does not want to invest the time and money in buying, configuring and managing VoIP equipment can opt for an IP Centrex service, the third type of VoIP. Here they will rent equipment, calling time and access to call functions from a network operator. While the provider will take responsibility for managing voice quality it may be more expensive in the long run.

IP Centrex is increasingly being targeted at smaller businesses that are prepared to accept a more limited range of features in exchange for lower up-front costs. Smaller businesses might also be more prepared to use consumer-grade software applications such as Skype. These allow users to make free calls to one another from their computer over the public internet - often called internet telephony. Obviously, no guarantees for call quality can be given but that is not stopping millions of people from using it.

Internet telephony has been with us for ten years or more but great strides are being made in call quality to the point where firms like Skype and Vonage can break calls out of the internet to terminate at ordinary telephones and charge for it.

So there you have it. IP telephony, in all its flavours.