Because of the time-sensitive nature of IP telephony applications, IP telephony should be implemented on an entirely switched network. Ethernet collisions, which are a major contributor to delay and jitter, are virtually eliminated on switched networks. Additionally, the procr, media server, and IP telephones should be placed on a separate subnetwork or VLAN (that is, separated from other non-IP telephony hosts). This separation provides for a cleaner design where IP telephony hosts are not subjected to broadcasts from other hosts and where troubleshooting is simplified. This separation also provides a routed boundary between the IP telephony segments and the rest of the enterprise network, where restrictions can be placed to prevent unwanted traffic from crossing the boundary. When personal computers are attached to IP telephones, the uplink to the Ethernet switch should be a 100 Mbps link or greater, so that there is more bandwidth to be shared between the telephone and the computer.

Avaya solutions for large flat subnets with thousands of devices on them is not a supported configuration. If IP telephones and Avaya servers will share a subnetwork with other hosts, the IP telephones and Avaya servers should be placed on a subnetwork of manageable size (24-bit subnet mask or larger, with 254 hosts or less), with as low a rate of broadcasts as possible. With this situation, a worst-case example is the scenario where IP telephones and Avaya servers are deployed on a large subnetwork that is running IPX or other broadcast-intensive protocol, with broadcasts approaching 500 per second. There is an arp cache limit of 1024. When the arp cache is full, it will be unable to communicate with any new hosts until the arp cache times out on other hosts. So, network segregation into smaller subnets like /24 or the creation of VLANs, or doing both is strongly recommended.