TMCnet
ITEXPO begins in:   New Coverage :  Asterisk  |  Fax Software  |  SIP Phones  |  Small Cells
 
May 2007
Volume 10 / Number 5
Feature Articles
 

Delivering Qos for End-to-End
VoIP Service Quality

By Jay Malin, Ph.D., Feature Articles
 

Advertising buzz terms for voice over IP such as “it’s the network” and the “sound of a pin drop” underscore the public’s interest in voice call quality. One provider of over-the-top VoIP services even claimed recently to provide “business class” quality.

While such an achievement is admirable, it is nearly impossible to make such claims regarding QoS without the ability to manage the subscriber’s broadband connection. At a minimum, the entire network must be engineered for end-to-end quality of service for the application being provided.

Quality of service for advanced IP applications can be measured first by the subscriber’s experience with the application (be it voice, video, or gaming) and then by the application’s specific transport flow specifications such as latency, jitter, and bandwidth. For example, for a high-quality VoIP phone call, the latency and jitter must be tightly controlled (approximately < 1/100 second) so that VoIP data packets can be appropriately reassembled by the application. For real-time video such as IPTV, sufficient bandwidth is required to transmit the data packets, and the jitter must be less than the buffer capacity.




Table 1 shows examples of applications and their associated resource requirement (e.g., flow specifications).

Taking this concept one step further, the intention is to quantify the quality of subscriber experience - called the QoE - and correlate the result to the necessary flow specifications. In the world of VoIP, the QoE is oftentimes defined by a mean opinion score (MOS), a number from 1 to 5 determined by surveys of actual human interaction with the media. Today, numerous tools allow service providers and subscribers to electronically determine the quality of a broadband connection via its associated impact on MOS.

In addition to the MOS score, subscribers are sensitive to service uptime and busy signals, which also contribute to quality of service. Service uptime is a reflection of reliability: The standard is called five 9s, or 99.999% and is best defined by the availability of network elements such as routers and softswitches. Therefore, while an over-the-top VoIP service provider may be able to guarantee the reliability of its own softswitch, it cannot make any commitments regarding the availability of the broadband network’s routers.

On the other hand, while a busy signal may indicate that the network is down, it may also be a good sign that the service provider is ensuring call quality by blocking calls when network capacity is reached. Service providers design networks with the Erlang traffic engineering model in mind to comply with their subscriber service level agreements (SLAs). The Erlang model accounts for available network capacity and the resources required for each call (as previously defined by MOS). The SLA can be managed either one of two ways: by over-engineering all the network resources, and/or by dynamically enforcing the SLA using a policy-based approach to network management.

This returns us to this article’s original premise that it is nearly impossible to make QoS claims without the ability to manage the subscriber’s broadband connection. The ability to over-engineer the network or apply a policybased approach to manage SLAs is available only to a VoIP or content provider with access to the network. One other approach - albeit lacking a service guarantee - is the implementation of a more robust audio and video CODEC that minimizes the network resources required to deliver the service. For the service provider, the choice comes down to cost.

In the case of the over-engineered network, the cost is raw broadband bandwidth, which for an MSO means splitting a node or, in a wireless network, adding a sector (or spectrum). Sometimes, this can be performed easily; however, the cost, in a world of proliferating bandwidth usage - is prohibitively expensive. Adding capacity without reserving resources for walled garden applications means that service providers who “own” the network not only cannot guarantee the QoE, but also cannot competitively monetize the value of their plant upgrades because the “all you can eat” bandwidth is available even for over-the-top content. On the other hand, by dynamically managing the allocation of network resources for premium advanced IP services, service providers can both monetize the value of the premium service and most efficiently exploit their capital plant. Policy management combines admission control and real-time, parameterized resource allocation in order to manage the delivery of IP applications such as VoIP (define - news - alert) and video. The figure below shows how a policy management architecture can be implemented for a broadband VoIP service. The admission controller enforces the traffic engineering model by comparing the number of concurrent application sessions against the network’s capacity. If the resources required to ensure the application’s performance comply with the SLA, the application (e.g., voice call) is blocked. This function can be reasonably performed in most IP networks. (See Figure 1.)

While many networks are being upgraded to add real-time bandwidth (and latency and jitter) allocation capabilities, cable and WiMAX stand out above the rest. The standards committees overseeing these networks have incorporated dynamic, parameterized QoS capabilities into their specifications. This means that when an application such as a voice call is admitted by the admission controller, the precise resources are allocated at the applicable router regardless of whether it is a CMTS or base station. Application-specific data packets are scheduled and handled in the network routers based upon both priority and resource requirements. Similar types of support are being added to DSL aggregation routers, WiFi access points, and 3G packet gateways.

In an alternate, less precise approach, data packets are marked by network elements for prioritization throughout the network. As long as it is combined with an admission control function and the markings are coordinated across all applications, the SLA can be reasonably enforced. However, because any packet can be marked by virtually any subscriber, this approach is not completely trusted and is not nearly scalable as the former approach.

QoS is most often addressed in the access network due to the competition for network resources. In the access network, many subscribers contend for limited network resources, thereby causing the congestion predominantly responsible for poor QoS. While the backbone is not completely immune to resource limitations, it is over-engineered to accommodate significant traffic. Network management technologies such as MPLS and DiffServ can be mapped to access network QoS policies to facilitate an end-to-end managed service.

End-to-end quality of service can be delivered only when the application provider has access to the network. In order to reliably deliver a high-quality IP service such as VoIP, service providers must incorporate an adequate traffic engineering model into a policy-based admission control engine. By offering a premium delivery service to subscribers who cannot be otherwise provided by a third party, service providers can monetize the value of their network while managing subscriber satisfaction.

Jay Malin, Ph.D., is VP of Business Development for CableMatrix Technologies, Inc., (news - alert) a leading provider of QoS policy management solutions for the broadband industry. For more information, please visit the company online at http://www.cablematrix.com.

 




Today @ TMC
Upcoming Events
ITEXPO West 2012
October 2- 5, 2012
The Austin Convention Center
Austin, Texas
MSPWorld
The World's Premier Managed Services and Cloud Computing Event
Click for Dates and Locations
Mobility Tech Conference & Expo
October 3- 5, 2012
The Austin Convention Center
Austin, Texas
Cloud Communications Summit
October 3- 5, 2012
The Austin Convention Center
Austin, Texas