February 2002

BY LAURA HOLLY

Session Initiation Protocol, or SIP, is the current darling in the Voice over Internet Protocol (VoIP) signaling protocol war, and seems to be the heir apparent to the incumbent H.323. In place of H.323�s complicated specifications, SIP is a text-based protocol (like HTTP) and offers a faster, simpler, extensible, and more scalable protocol for initiating, modifying, and terminating call sessions. It�s no wonder enterprises and service providers are finding SIP to be an attractive alternative.

But perhaps most compelling is the potential of SIP as an enabler of next-generation, high-value services. Beyond today�s enhanced services, like call forwarding or do not disturb, SIP offers the promise of an explosion of new services, such as unified messaging, instant conferencing, and a host of other services integrating voice with Web, e-mail, instant messaging, and other presence applications.

The QoS Challenge
If, however, SIP is the big kick that VoIP has needed in order to displace the enterprise PBX and bring an IP phone to every desktop, then why in 2002 are we still waiting for the heady VoIP revenues the industry has been predicting every year for the last several years?

Ask enterprises why they have not yet deployed VoIP in production, and they will tell you it�s because they have concerns about the quality and reliability of such a service. Regardless of the number of new services, SIP and VoIP applications won�t truly achieve ubiquity until users are convinced that they are consistently receiving the same level of quality they get from the PSTN. Quality in a SIP-based VoIP network is in the eye of the beholder and must reflect the experience of the user. However, in this quality equation, there are a number of different �users:�

• VoIP retailers purchasing minutes from an international wholesaler need to be sure that their retail service subscribers are receiving quality service;
• Traditional telephony providers off-loading excess PSTN minutes onto a VoIP network need to make sure their customers receive the toll quality they expect;
• VoIP wholesalers running minutes through multiple peering partners need to evaluate the quality offered by these partners;
• VoIP network providers need to understand the effectiveness and QoS impact of least cost call routing policies; and
• Enterprises need to evaluate whether their networks can support and sustain voice applications on a converged voice and data network.

With so many kinds of users with different perspectives on what the service is, that�s a tall order. However, as a call control protocol SIP, by itself, offers no QoS mechanisms, such as network resource reservation or best quality routing.

A New Approach
Both service providers and enterprises are finding the solution to this critical quality challenge in a new approach to service performance verification -- one that enables true, end-to-end and edge-to-edge quality measurement for VoIP services from the service user�s perspective. What sets this new approach apart from previous network monitoring schemes is its comprehensive scope, ensuring performance verification throughout:

• The entire VoIP Infrastructure -- registration, redirect and proxy servers, firewalls, naming infrastructure, gateways, intelligent SIP end-points, and the intervening network from many testing locations;
• The entire call session -- from call setup to media transfer to call termination; and
• The entire service lifecycle -- from pre-deployment testing to production network monitoring, testing, and troubleshooting.

The result is a carrier-class, real-time VoIP performance and call quality measurement system that provides pervasive, end-to-end service verification. The verification system is composed of three main elements:

• Verifiers -- These distributed, purpose-built hardware platforms are deployed throughout the network from customer premise to POP to peering point, and anywhere that ownership or responsibility changes hands, to create service demarcs. These verifiers actively test by initiating SIP-based transactions and calls and passively test by monitoring customer-generated traffic for performance.
• Verification Server -- A centralized software platform provides configuration and management for the distributed verifiers, continuous auditing of performance data against user-specified quality criteria, and a web-based Service Portal to view results and configure reports.
• Voice Service Suites -- Verifiers execute a wide array of highly configurable test modules to evaluate the performance of VoIP networks, infrastructure, and applications. These tests include SIP call signaling, RTP-based media path, and voice quality, plus performance tests for important ancillary services such as DNS.

Comprehensive VoIP Testing
By its very nature, the quality of real-time, two-way voice communication is susceptible to packet network phenomenon such as delay and loss. In data applications, normal network delay is of little consequence and lost packets can simply be re-transmitted. Not so for voice. For example, humans are very sensitive to delay in voice conversations and will find one-way delays of more than 150ms noticeable and on the order of 400ms down right unacceptable. In voice applications, delay manifests itself as echo or talk overlaps, while packet loss, especially consecutive bursts of loss, will result in clipped speech or dropout. Clearly, the margin of error for voice quality is small.

While round trip delay is an important measurement, it is not sufficient. Asynchronous routing of calls through the voice network makes it impossible to divide a round trip delay measurement by two to determine experienced latency -- if the delay from A to B is 100ms, but the delay from B to A is 400ms, a reasonable 250ms average delay does not reflect the annoying call that A and B will experience. Only by examining one-way delay values will a provider have an accurate reflection of the real latency impact.

Jitter, or variability in the inter-arrival time between VoIP packets, is typically compensated by jitter buffers that smooth the replaying of an audio stream. However, excessive jitter can require larger jitter buffers which cause longer delay and finally loss as packets are discarded from the jitter buffer to make room for newly arriving packets.

While delay, packet loss, and jitter are crucial indicators of VoIP network performance, what network operators really want is a single measurement of user-perceived call quality to easily verify that they are delivering toll quality service on every call. The ITU�s E-Model (G.107) is well suited to this task, computing a 1-100 R-factor score reflecting the delay and loss experienced on a call. R scores can easily be converted to a traditional MOS (mean opinion score). Unlike computationally intensive perceptual models like PAMS and PSQM, which can take many seconds to compute, E-Model-based quality measurements can be computed in real time, to provide continuous or high-frequency quality monitoring in large, operational networks.

For VoIP applications, it is just as important to measure call setup as well as the media session performance because both components affect quality of service. SIP-based call setup involves multiple cooperating proxy, registration, and redirect servers, all of which much be tested.

For SIP, important session measurements should include:

• Post dial delay (PDD) -- ring confirmation after initiating the call;
• Total call setup -- the time from initiation until call is accepted (SIP INVITE -> OK);
• First media receipt -- the time to receive the first media packet;
• Registration -- the time required to complete the client�s registration with the SIP registration server;
• Authentication -- the time required to complete client�s authentication with theSIP proxy server;
• Redirection -- the time attributable to redirects, e.g., when a user has moved temporarily;
• OPTIONS -- the time required to complete capability negotiation;
• SIP status completion code -- was the call completed successfully or not; and
• Call termination time, the time to terminate the call (SIP BYE -> OK).

Testing Methodology
To gather these measurements, a verifier emulates the behavior of a SIP user agent client against either another verifier acting as a SIP user agent server, or a SIP endpoint (e.g., proxy server) by generating actual SIP call setup directives and establishing an RTP-based media stream. There are benefits to both approaches. Testing between two verifiers offers highly accurate results, especially for one-way delay. Packet time stamping performed in the verifier hardware and NTP or GPS-based clock synchronization support microsecond timing accuracy.

Verifiers can also be easily deployed in various locations to test complex mesh, hub and spoke, and other network configurations. On the other hand, testing against an actual endpoint gives direct feedback about how well that endpoint is performing in the VoIP network.

In order to completely test a VoIP infrastructure, both active and passive testing is also recommended. Active testing offers fine grain control over all aspects of testing, such as codec specification, length of the call, frequency of testing, and specification of originating and terminating location, to simulate a variety of real-world usage patterns. Because active testing can be done independently of users, performance measurements can be collected 24/7 and used to provide proactive identification of quality issues before users are affected. Important call completion and error call ratios (such as ASR, NER, etc.) can be calculated equally well from active and passive testing data. Passive testing, which monitors actual customer-generated calls, also offers indirect quality information about average call duration (ACD).

The Ultimate Test
As innovative technologies like SIP emerge to simplify development and delivery of VoIP services, the barriers to market entry will continue to fall. While that�s good news for service providers and enterprises looking to reap the benefits of VoIP, QoS concerns will continue to hold down deployment rates.

Monitoring a segment of the network or just VoIP media path performance is not nearly enough. To answer these quality concerns, providers and enterprises will need a scalable, performance verification system that comprehensively assesses call quality from one end of the service to the other, using actual VoIP protocols and delivering voice quality results in real time. Given today�s increasingly competitive environment, deploying such a measurement system from the outset could be the key to passing the ultimate test: Attracting and retaining subscribers.

Laura Holly is product line manager at Brix Networks, a Chelmsford, MA-based provider of IP service assurance and performance-monitoring solutions. Please visit their Web site at www.brixnet.com.

[ Return To The February 2002 Table Of Contents ]