Dodging VOIP Predeployment Pitfalls

Case Study: Companies tackle demands VOIP rollouts place on networks.

Adding voice onto an existing IP network is not the same as adding any other data application. Successful VOIP rollouts require careful planning, proper design and—in most cases—changes to the networking infrastructure to adequately support latency-sensitive voice traffic.

But the first and most critical step that enterprises can take to ensure a successful voice-over-IP installation is a predeployment assessment of the network that will carry the voice traffic.

Fully 85 percent of existing networks cant handle VOIP traffic, and 75 percent of enterprises that dont analyze their IP network infrastructure before deployment wont see a successful VOIP implementation, according to Gartner Inc., of Stamford, Conn.

/zimages/6/28571.gifClick here to read about why some VOIP operators arent dialing in to 911.

To handle the increased demands that voice places on data networks, "90 percent of networks in North America today will require additional build-out to support voice, and 100 percent of them will require some configuration changes," said Gartner analyst Jeff Snyder.

Beyond the obvious changes required—such as upgrading slow LAN or WAN links, and adding POE (power over Ethernet) to network switches—there is a host of network problems that often are not seen by data applications or are tolerated because of their infrequency. No matter how good network engineers believe the network is, such problems can derail a VOIP implementation.

"IP networks by design are just good enough," said Loki Jorgenson, chief scientist at Apparent Networks Inc., a Vancouver, British Columbia, company that supplies VARs and systems integrators with a tool that can find such problems. "Things that have been a nuisance for network engineers are critical tripping points for voice. Networks are riddled with potential degradations that are often hard to find but easy to resolve. So you have to start out with the assumption that your network is dirty, and you have to bring it up to an acceptable level for voice."

Jorgenson said he believes there is "an ecology of Layer 1, 2 and 3 defects" that can derail VOIP deployments. A predeployment assessment using tools such as Apparent Networks AppareNet Voice can uncover NIC (network interface card) driver issues such as full- and half-duplex conflicts.

When ISPs implement rate-limiting queues that throttle predefined traffic groups when bandwidth is in contention in their Cisco Systems Inc. routers, packets can be lost, and voice quality suffers, according to Jorgenson.

In addition, there is always latency that can be caused by myriad contributors.

"Satellites introduce 250 milliseconds of delay," Jorgenson said. "Voice should always be below 150 milliseconds. Devices being added in the path can push you over a threshold. I dont think theres a lot of awareness of latency at this level, and that makes it hard to make good design choices."

Then there are media errors that can be caused by "anything from cabling to bad optic ends, electromagnetic interference, wireless random loss or corruption," Jorgenson said. "Media errors are extremely difficult to detect and resolve, and they will show up as jitter variation and loss and have no impact on any other application except voice."

Using the AppareNet Voice tool, engineers at managed services provider IPC Technologies Inc., in Richmond, Va., start their predeployment assessments by testing a customers WAN links.

"Youll see call-quality issues there first," said Kurt Wright, senior network engineer at IPC. Engineers execute tests that measure bandwidth utilization, packet loss, round-trip time, latency and packet reordering. They also look at mean Opinion scores, which measure call quality.

Next Page: Asking the right questions to prevent disaster.