Tech Scan

December 5, 2002
Quality of Service (QoS): A Requirement for IP Telephony?

Vijay Yadav

Over the past decade, traffic on Internet has increased significantly along with a multitude of applications ranging from streaming video, multimedia, e-commerce and VPNs. Although many bemoan the slow roll out of high-speed broadband access technologies such as ADSL and cable, but the figures are quite impressive. Over 40 percent of Internet households have access to broadband connections in countries such as Germany and Switzerland, according to Nielsen.

Furthermore, there are growing ranks of service providers who are offering low cost voice calls to distant countries using the Internet for the majority of the call path. More traditional voice carriers are gradually introducing an expanded menu of voice offerings, which utilise IP transport to some degree, such as PC-to-phone calls, IP-hosted telephony, voice portals and advanced directory services using voice recognition.

While voice communication is possible over 'best effort' delivery methods provided by the Internet, many believe comprehensive network upgrades will be required to provide the necessary QoS for end-to-end voice calls. This is by no means a universal view.

Whereas, IP telephony users who regularly use PDAs, PCs and IP phones to make calls are reporting excellent results with devices connected to both fixed and wireless LAN networks. Empirical evidence seems to suggest that as long as each user participating in a voice call or voice conference has access to a service provider with sufficient excess capacity, performance is similar to the regular telephone network.

This has lead to two schools of thought on how to enhance and modify IP networks for voice usage. The first is to throw bandwidth at the problem and make sure that there is sufficient diversity of connections and the congestion is infrequent. The other approach is, of course, to engineer the packet network so that it recognises and prioritises voice traffic over other types of traffic and ensures that quality can be guaranteed under all circumstances. The traditionalists who are accustomed to a circuit being dedicated to a call for its duration tend to prefer the latter approach, while those who have pioneered the use of IP for voice often promote the former.

What prevails will depend on the market and the carriers who are taking different approaches in different markets. Some carriers are already using VoIP in the core of their networks without the customers having any idea that the call is being packetised. These carriers have fully addressed the network issues, often in the simplest way - dedicating IP capacity to VoIP and not trying to mix voice and other traffic on the same circuit. Using this approach the network engineering becomes relatively simple as the bandwidth required can be easily calculated on the maximum call capacity for the trunk.

Over time the market will decide whether premium price or standard performance will triumph in the end. Successful service providers will be those who innovate by creating new services that attract customers and simultaneously delivering the services with the quality expected by the users.

Measuring voice quality
Users expect VoIP quality and network reliability to be the same as for traditional telephone networks and the benchmark for this is PSTN. There are a number of ways in which you can quantify voice quality, but probably one of the most meaningful is the mean opinion score (MOS). It rates calls on a scale of 1 to 5. This is solicited from groups of real people who use a connection and rate the quality of a call.

It might be assumed that the result would vary widely, with a large enough samples of testers consisting of people. If the resulting MOS rating is above 4, then the connection is said to be 'toll grade', or as good as the PSTN.

Gathering a group of testers together at a moment's notice is hardly a practical way to assess the quality of a given connection, therefore alternative methods such as the International Telecommunications Union (ITU) E-Model have emerged. This model enables voice engineers to estimate the likely quality of a connection under a given set of impairments. The output of the E-Model is a 'transmission rating factor' called the R-Value, which can vary between 0 and 100. This can be mapped to a MOS with R-Values greater than 80 being roughly equivalent to toll grade.

Facing the issues
Various challenges emerge when voice is introduced onto data oriented packet networks. Transmission conditions that pose little threat to data traffic can introduce severe problems when it comes to packetised voice. To ensure toll quality connections, the impact of all of the following have to be understood and controlled to deliver an appropriate MOS or R-Value:

q Real-time Bandwidth - While data can tolerate packet loss, delay and can make use of widely varying bandwidth, voice needs constantly available bandwidth. Although compression reduces the bandwidth required, most calls will require a certain minimum bit rate to be supported end-to-end.

q Compression - Compression is widely used in packet networks to reduce the bandwidth required. Codecs (compression coder/decoders) that achieve 2:1 or 4:1 compression can still produce very good MOS, however significant impairment can occur when a call gets compressed and decompressed more than once, therefore this should be avoided where possible.

q Packet Loss - IP networks can lose packets for various reasons. When networks get congested voice packets may get delayed and arrive too late to form part of the reconstituted audio. Routers that are managing congestion may drop other packets. Although some loss can be endured, most people will not tolerate lost syllables or even whole words during conversations.

q Delay - There are many sources of delay, including codec processing delay, packetisation delay, queuing/buffering delay, network switching delay etc. If the end-to-end delay along the call path becomes too long, then communication becomes difficult. Up to 150 milliseconds delay is generally considered acceptable.

q Jitter - Jitter is another factor that affects delay. Jitter occurs when there is a variation between the expected arrival of a voice packet and when it is actually received, causing a discontinuity in the real-time voice stream. Jitter buffers can reduce this impact, but they introduce delay.

q Echo - When uncontrolled echo occurs on a voice call it can make speaking quite difficult if it is returning to the speaker's ear with more than approximately 25 milliseconds delay. In voice packet networks, echo cancellers are normally used to combat this issue.

End-to-end quality assurance
The overall objective of QoS is to optimise bandwidth, improve throughput, and reduce latency, jitter and packet loss. Adding more bandwidth by over provisioning individual links is one way of addressing QoS problems but may not be cost effective. Service providers need to look at the economics of their own particular environment and decide whether additional bandwidth or a more sophisticated approach is required, which will increase the cost of many components in the network.

Packet classification, prioritisation, congestion management, and traffic shaping and policing are the mechanisms that can be applied to the network traffic to ensure end-to-end QoS. But many carriers will not only be handling calls within their own network; they also need to hand them off to other carriers. Therefore any solutions for service providers must be scaleable and ensure that end-to-end QoS can also be provided for calls that transit multiple carriers.

Two important techniques worth delving into are DiffServ (RFC 3260) and multiprotocol label switching (MPLS). DiffServ is a major model for providing QoS with differential pricing developed by the Internet engineering task force (IETF) as a scalable solution for service providers.

DiffServ is an alternative to a more resource-intensive model called Intserv, which relies on RSVP (resource reservation protocol) to set up and tear down resources along a reserved path. DiffServ eliminates the need to use RSVP and instead provides for a much simpler scheme. It enables service providers to provide different service characteristics to all the traffic requesting that service profile. This results in a coarser approach, but one that scales much more easily in large networks. The network consequently is not trying to allocate resources to each individual session requesting QoS, but groups traffic that can then be allocated particular resources delivering a specific class of service (CoS). DiffServ is applied at the edge of the network and packets are marked so that routers can handle them appropriately.

Perhaps the most touted technology for enabling service providers to offer QoS is MPLS. MPLS is essentially a mechanism by which a pre-arranged route is set up through the network to handle a given flow of packets. Unlike DiffServ, which operates at the IP layer, MPLS assists routers to map IP flows to layer-2 transports like frame relay and ATM. This technology enables routers to switch traffic without having to do a route look up based on the IP address. Edge routers examine the IP address and any QoS requirements for traffic and then label it such that each MPLS router simply switches to the appropriate output with minimum processing. By allocating traffic according to MPLS paths different service levels can be provided.

Significance in business
These standards, which are now available for addressing QoS in voice-over-packet networks, allow carriers to offer service level agreements (SLAs) with meaningful voice-quality guarantees that current and potential users demand.

It also enables the carriers to price the services according to agreed-upon levels, thus creating differentiated offerings and perceived value for higher-quality, more expensive services. SLAs can be implemented in various ways, including:

q Assigning higher priority levels to 'gold member' subscribers

q Reserving a fixed amount of bandwidth for mission-critical applications such as voice and video

q During heavy periods of congestion, designating which traffic should be dropped. Also, controlling and limiting the amount of outgoing bandwidth

q Offering new value-added services, which are based on mission-critical applications and customers requiring 'always-on' guaranteed connections for bandwidth-intensive applications

Looking ahead - IP telephony benefits
QoS over the Internet has far-reaching effects beyond just voice, including high-bandwidth video and multimedia information. A few years ago, industry experts predicted that IP telephony would be deployed fast and furiously. However, service providers were not so quick to adopt this technology. One of the big barriers was lack of voice quality. But the situation has changed considerably in recent times.

Technological advances, methodological approaches to QoS in voice-over-packet services and experiences have proved that VoIP is quite capable of meeting, and indeed exceeding, the quality of circuit voice.

A growing market for IP Telephony is expected to continue throughout the first half of the decade. For example, Ovum Ltd. predicts that the number of active PC-to-phone users will rise by more than ten-fold, from 11.3 million in 2001 to 166 million in 2006. IP telephony traffic will explode from 3.4 billion minutes in 2000 to 251 billion in 2005 and PC-to-phone service revenue will increase, from $185 million in 2000 to $6.2 billion in 2005.

Due to IP telephony's strengths, such as enhanced service generation, rapid deployment and a cost-saving architecture, many service providers can now enter the voice market.

ISPs now have the means to offer voice calling; retail Internet telephony service providers (ITSPs) can offer competitively priced phone-to-phone and PC-to-phone IP telephony services to consumers and businesses. Application service providers (ASPs) can deliver wholesale IP telephony services to portals and ISPs and wholesale ITSPs can provide service via the Internet to other carriers. An entirely new market has opened for IP telephony clearing houses to manage traffic exchange and financial clearing between ITSPs.

In addition, traditional carriers now have much more cost-effective means to deliver IP-based voice services in global areas that their own network does not reach, or where they require additional capacity.

With competition to attract consumer, service providers need a powerful vehicle to boost their subscriber base and build their revenues. With trust building in packetised voice, IP telephony and more, giving providers a means for lowering infrastructure costs and introducing enhanced new services as they strive to become increasingly profitable.

(The author is country manager, CommWorks, India)