The plan was very simple and straightforward. The goal was to provide QoS for the voice traffic end-to-end across the IP LAN via a DiffServ model, using Class of Service (CoS) and Differentiated Services Code Point (DSCP) where appropriate. DSCP refers to how the network traffic is marked or tagged in the IP layer (Layer 3) to identify it as part of a particular class. Voice packets are marked with DSCP (DiffServ code point) on the end device; this marking is mapped to a CoS tag in the LAN switch. CoS refers to categories of network traffic that are recognized and queued by the network switches. CoS marking or tagging occurs in the Layer 2 header for the 802.1q trunks, and this method is referred to as 802.1p DiffServ, as the overall model defines how these packets are queued and forwarded in the network. Based on the policies set, network switches use CoS and DSCP values to provide high-priority handling for voice traffic.

The life of an audio packet as it is transported over an IP network is given here:

• Human speaks
• Coded with some form of compression
• Inserted into packets with sequence numbers
• Transported onto the network
• Received into network equipment
• Network determines next-hop interface and switches the packet
• Packet is queued for sending to the next hop
• Received in a playout buffer
• Decoded in sequential order
• Played back to human

As packets traverse the network, all elements must recognize which packets are real-time audio and prioritize their handling, yet ensure that no uplink, or any other portion of the network, becomes over utilized by the addition of voice on the network.

Another factor is the coder-decoder or codec used to compress and decompress voice transmission. G.711 is a standard for speech codecs that provides toll quality audio at 64 Kbps. Intel elected to use uncompressed G.711 as a codec in this trial since it provides the highest quality available for the system implemented, and bandwidth utilization is not an issue on the LAN.