Second Quarter 1998

BY PRABHU KAVI

As time-sensitive voice and video applications migrate to the Internet, the network infrastructure must evolve to meet transmission needs that are markedly different from those of conventional data applications.  In this evolution, the ability to support demanding toll-quality voice over IP (VoIP) applications will be a key success factor for emerging voice service providers.

QoS REQUIREMENTS FOR VOICE
Before exploring the changes that must occur at the core of the Internet infrastructure, it is helpful to establish the three requirements for toll-quality voice communications.

Throughput
Telephone calls require a near constant bandwidth rate. In the Public Switched Telephone Network (PSTN), uncompressed voice requires 64 Kbps in each direction, even when one person is silent. By compressing the digitized voice, the bandwidth requirement drops to about 8 Kbps, excluding silence suppression and IP overhead. Unlike applications such as file transfer that will speed up to use excess network bandwidth, voice calls generate the same traffic even when excess bandwidth is available. Conversely, voice calls do not adapt well to substantial packet loss that can occur when the available end-to-end bandwidth drops below the desired rate (unlike file transfer applications that adapt to whatever bandwidth is available).

Latency or Delay
Anyone who has experienced a telephone call via satellite can appreciate the need for minimal latency in a telephone call. A single-hop satellite link adds about a quarter of a second one-way delay. This seemingly small delay is annoying in a conversation, and is the upper limit for acceptable latency.

Delay Variance
Because the Internet has variable delay, VoIP gateways must incorporate large buffers for smoothing the various packet delays into a constant delay voice stream. The need for these large buffers further increases the end-to-end delay beyond what the network layer imposes.

SOLUTIONS FOR TOLL-QUALITY
It is obvious that a solution for toll-quality voice over IP should strive to guarantee bandwidth, minimize latency, and minimize delay variance. The Quality of Service (QoS) mechanisms that are becoming available in the Internet now can be explored relative to the requirements of voice traffic. 

"Best Effort" QoS describes the current state of the Internet. The routers in the Internet make no guarantees about throughput for any voice call, packet loss, or latency. The underlying problem is that routers treat all packets the same, regardless of whether they’re from a time-sensitive voice call or a delay-tolerant file transfer. A best-effort infrastructure cannot properly support voice traffic.

Relative QoS prioritizes traffic using the Type of Service (ToS) byte in the IP header. This byte, long unused, is being redefined to represent the packet’s delay priority and drop priority. Equipment vendors are starting to adapt their products to support the appropriate priority queuing and congestion control mechanisms. VoIP gateways need to tag packets to request low delay and low likelihood of being dropped. Relative QoS is a substantial improvement over Best Effort, and can yield acceptable voice over IP quality in a well-managed IP network. However, the need for the network operations and engineering staff to monitor the usage of voice traffic on each trunk adds a substantial amount of unnecessary work, considering that better alternatives are available.

Absolute QoS builds on Relative QoS, but also guarantees sufficient bandwidth and bounds on the latency and delay variance. Absolute QoS is necessary for taking Internet telephony from the experimental stage to mission-critical production use. 

THE CORE: WHERE THE ACTION IS
Many large Internet networks are already built on top of a frame relay or ATM core. Service providers initially built their Layer 3 connectionless IP routing network on top of these Layer 2 connection-oriented technologies to take advantage of the higher throughputs that ATM and frame relay switching equipment had when compared to routing equipment.

The line between Layer 3 routing equipment and Layer 2 switching equipment is beginning to blur with the availability of software that brings IP routing functionality to ATM and frame relay switches. This approach combines the intelligence of IP routing with ATM and frame relay QoS. This combination enables toll-quality voice over IP for the first time.

Figure 2 illustrates how voice over IP gateways and an Absolute QoS-aware IP core network interact. The voice over IP gateway would set the IP header’s Type of Service byte to a specific value for voice packets (e.g., low latency and low drop probability). The core network looks for IP packets with this specific ToS value, and when the first such packets are seen, sets up a Switched Virtual Circuit (SVC) between the source VoIP gateway and the destination VoIP gateway with bounds for latency, delay variance, and reserved bandwidth. Subsequent voice packets belonging to the same call are automatically sent over this established SVC, and the SVC is torn down when call traffic stops.

Quality Is In The Switch
Sending voice packets over an SVC is analogous to how quality is guaranteed in the PSTN, and is feasible over an ATM or frame relay core if the switches being used in that core have SVC setup rates rivaling the call setup rates of today’s largest voice switches.

Networks that use IP routing at the edge, and either ATM or frame relay in the core, have some substantial advantages over existing voice switches. These advantages include:

• Equipment Cost: Switches that combine IP routing with frame relay and ATM QoS are about the same cost as IP-only routers, and are a fraction of the cost of voice switches.
• Bandwidth Cost: Unlike voice switches that dedicate 64 Kbps in each direction, Absolute QoS-aware IP networks only reserve the amount of band-width needed for a compressed VoIP stream (typically 13-16 Kbps). Since the bandwidth is only reserved, and not dedicated, other traffic classes (e.g., file transfer) can use any leftover band-width, as happens when silence suppression detects one party is silent. This bandwidth conservation allows the service provider substantial savings in monthly bandwidth cost.
• Operations Cost: Networks using Absolute QoS-aware equipment can support Relative QoS and Best Effort IP traffic in addition to Absolute QoS IP traffic. Since the network uses either a frame relay or ATM core, these services can be offered to customers as well. The ability of a single network to offer these services under a single network management system yields substantial operational cost savings over building separate networks to offer these services.

WHAT’S NEXT?
VoIP service providers that guarantee Absolute QoS have a strong competitive advantage — the ability to support voice applications with the clarity and performance of today’s toll calls. When end-to-end networks are Absolute QoS-capable, a VoIP call can assure toll quality across any number of IP networks. Without Absolute QoS, no guarantees are possible. In these cases, VoIP service providers should connect to IP networks that support at least Relative QoS in order to offer some, albeit unpredictable, performance priority over ordinary data traffic.

The upgrading of the entire Internet will take many years. In the interim, VoIP service providers should peer with other service providers who understand the importance of Quality of Service.

Prabhu Kavi is IP Business Marketing Manager, Core Systems Division, Ascend Communications, Inc. Ascend is a leading provider of technology and equipment solutions for telecommunications carriers, Internet service providers (ISPs), and corporate customers worldwide. Ascend delivers a comprehensive set of best-of-breed solutions in the key areas required to build a high-performance, cost- effective public and private network infrastructure from end to end. For more information, contact the company at 510-769-6001 or visit their Web site at www.ascend.com.