Industry best practices dictate that a network be designed considering the following factors:
Voice mandates consideration of the following additional factors when designing a network:
In general, these concerns dictate a hierarchical network that consists of at most three layers (Layers in a hierarchical network):
In some networks, a single device can perform the functions of several layers.
Table 1: Layers in a hierarchical network
Layer |
Description |
Core |
The core layer is the heart of the network. The core layer forwards packets as quickly as possible. The core layer must be designed with high availability in mind. Usually, these high-availability features include redundant devices, redundant power supplies, redundant processors, and redundant links. Today, core interconnections increasingly use 10 Gigabit Ethernet or higher. |
Distribution |
The distribution layer links the access layer with the core. The distribution layer is where policy like the QoS feature and access lists are applied. Generally, Gigabit Ethernet connects to the core, and either Gigabit Ethernet or 100base-TX/FX links connect the access layer. Redundancy is important at this layer but not as important as in the core. This layer is combined with the core in smaller networks. |
Access |
The access layer connects servers and workstations. Switches at this layer are smaller, usually 24 to 48 ports. Desktop computers, workstations, access points, and servers are usually connected at 100 Mbps or 1 Gbps. Limited redundancy is used. Some QoS and security features can be implemented in the access layer. Mostly, Power over Ethernet (PoE) is included to power IP telephones and other access devices. |
For IP Telephony to work well, WAN links must be properly sized with sufficient bandwidth for voice and data traffic. Each voice call uses 9.6 kbps to 120 kbps, depending on the desired codec, payload size, and header compression used. Additional bandwidth might be used if video or redundancy for fax, modem, and TTY is implemented. The addition of video can stress WAN links engineered for voice only. WAN links must be re-engineered when video is introduced to an existing network. The G.729 compression algorithm, which uses about 27 kbps of bandwidth, is one of the most used standards today. Traditional telephone metrics, such as average call volume, peak volume, and average call length, can be used to size interoffice bandwidth demands. For more information, see Traffic engineering.
Quality of Service (QoS) also becomes increasingly important with WAN circuits. In this case, QoS means the classification and the prioritization of real-time traffic such as voice, video, or FoIP. Real-time traffic must be given absolute priority through the WAN. If links are not properly sized or queuing strategies are not properly implemented, the quality and the timeliness of voice and data traffic will be less than optimal.
The following WAN technologies are commonly used with IP Telephony:
Multiprotocol Label Switching (MPLS)
Asynchronous Transfer Mode (ATM)
Frame Relay
Point-to-point (PPP) circuits
Internet VPNs
MPLS, ATM, Frame Relay, and PPP circuits, all have good throughput, low latency, and low jitter. MPLS and ATM have the added benefit of enhanced QoS. MPLS is a relatively new service offering and can have issues with momentary outages of 1 to 50 sec duration.
Frame Relay WAN circuits can be difficult to use with IP Telephony. Congestion in Frame Relay networks can cause frame loss, which can significantly degrade the quality of IP Telephony conversations. With Frame Relay, proper sizing of the committed information rate (CIR) is critical. In a Frame Relay network any traffic that exceeds the CIR is marked as discard eligible, and is discarded at the option of the carrier if it experiences congestion in its network. Because voice packets and other real-time packets must not be dropped during periods of congestion, CIR must be sized to maximum traffic usage. Also, Service Level Agreements (SLAs) must be established with the carrier to define maximum levels of delay and frame loss and remediation if the agreed-to levels are not met.
Internet VPNs are economical but more prone to quality issues than the other four technologies because there is no control or SLA to modify the handling of voice packets over data packets.
Network Management is another important area to consider when implementing IP Telephony. Because of the requirements imposed by IP Telephony, it is critical to have an end-to-end view of the network and ways to implement QoS policies globally. Products such as HP OpenView Network Node Manager, Prognosis, Concord NetHealth, and MRTG help administrators maintain acceptable service. Outsource companies are also available to assist other companies that do not have the resources to implement and maintain Network Management.