Provisioning QoS

QoS algorithms will help manage and monitor data-packet loads.

Subscriber services such as voice mail, fax services and voice messaging originally were delivered in the form of legacy systems. These systems resided on older standalone computers or on emerging high-speed servers. However, the wireless industry moved away from the legacy architecture with the development and availability of application-development tools implemented on UNIX or Windows NT operating systems. With these development tools, the CORBA systems emerged. These systems support the development of a new generation of "service ware." Perhaps one of the most novel systems to evolve from CORBA was over-the-air service provisioning (OTASP). This software package greatly simplifies the process for initializing new subscribers.

Service providers are able to enhance customer acquisition by allowing potential subscribers to buy off-the-shelf handsets and get instant OTASP.

Traffic & QoS
The rules for wireless-network service management are changing with the promise of 3G and high-speed data-packet traffic. These new networks promise greater bandwidth with an advancing array of Web-access choices. The promise of greater bandwidth will bring about the application of powerful portable computers and computer phones. Based on the rapid beta testing of GPRS, WAP and other related efforts, the coming generation of wireless networks will be burdened with a new level of bandwidth-intensive traffic. This traffic will be generated by a demand for personal multimedia applications. Wireless link characteristics, as well as the need to maintain connectivity while on the move, offer significant challenges for maintaining wireless quality of service (QoS).

End-to-end delay or bandwidth vary across a wide range for different classes of service. Therefore, efficient and effective QoS provisioning techniques are important in maintaining an active and satisfied network.

When engineering a wireless network, QoS satisfaction can be accomplished through the application and maintenance of recognized industry standards for configuring a wireless network. For example, in the design and engineering of a network, at its very base level, sufficient hardware resources should be made available to accommodate an overload of at least 30%. Further, bandwidth of a cell could be configured with provisions for either dynamic or static allocation of bandwidth.

By prioritizing voice traffic over data, voice traffic can be assured access to the network. These strategies coupled with complex, built-in routing algorithms and software for detecting network overloads allow voice traffic to flow evenly across the network. Although a standardized approach to network configuration may be sufficient for the typical circuit-switch wireless network, it may not apply in a 3G packetized network characterized by a high demand for more bandwidth.

Service providers are taking their networks that were engineered for voice and overlaying them with data networks. On top of this, they are introducing packetized data traffic that will support Internet access. This presents an entirely different set of problems relative to the management of traffic across the network to ensure QoS.

How best to deal with this depends on the availability of sophisticated software tools to help the network manager establish resources to maintain the desired level of QoS. To date, there have been a number of proposals for algorithms that could be used to support online QoS.

Algorithms for QoS
How best to maintain and track an acceptable level of QoS has been the basis for debate. Industry hype tends toward the predictability of QoS with little regard for the time between prediction and the avoidance of a network failure. Amid this debate, academic papers have argued the issue of QoS and addressed various solutions to the QoS problem in wireless networks. Some of these analytical models considered various levels of mixed class of service such as pedestrian and mobile, each with different types of channels and resource requirements. For example, handoff calls were given priority over other call traffic in these models. Probabilities were computed for performance characteristics for carried traffic, call blocking, forced termination and other related parameters for each traffic classification. Through these computed probabilities, you could estimate and in some cases predict call blocking, thus establishing a level of QoS.

Other QoS algorithms suggested a form of adaptive resource sharing among real-time traffic and non-real-time traffic. For example, real-time traffic would receive priority over non-real-time traffic. In this situation, real-time traffic can't tolerate delay while non-real-time traffic could tolerate some delay. A good example of this might be seen in voice IP packets vs. true data IP packets or 911 calls, which would supersede all priority levels for network traffic.

Another set of proposed QoS models is based on the concept of graceful degradation and guaranteed seamless service. The first of these proposals allocates all of the existing bandwidth to calls in progress, thus providing for a graceful degradation. Calls (data packets) that can tolerate delay would lose some bandwidth. The seamless proposal provides connectivity on the move based on minimum-requirements criteria.

Proposed Integrated Algorithm
Three researchers at the University of Texas (Das, Maink and Naveen) have proposed an approach that would integrate several of the above schemes into algorithms that provide an adjusted QoS based upon network conditions. They suggest that carried traffic in a wireless network can be decreased by graceful degradation of some or all of the existing services in a system. In this scenario, the quality of each connection deteriorates as data is discarded by the base-station transmitter to adjust to the reduced bandwidth.

The framework of this model provides for a differential treatment of real-time (delay-sensitive) and non-real-time (delay-tolerant) multimedia applications. In this setting, QoS provisioning involves:

• Development of efficient bandwidth-reservations schemes
• Sorting real-time vs. non-real-time packets
• Provision for packet marking
• Designing various priority scheduling techniques.

During call setup, the application would specify:

• Average bandwidth required
• Minimum bandwidth required
• Determination of whether the application is delay-sensitive or delay-tolerant.

In this situation, the application forwards its requirements profile comprising the above parameters. An admission controller then allocates one or more channels matched to the profile based upon traffic conditions. If bandwidth is available, it allocates to each non-real-time task on a time-shared basis.

Real-time traffic that cannot tolerate delay would be blocked or dropped if bandwidth isn't available. In some situations, a real-time call could force a non-real-time task to be buffered so that channel space could be released to the real-time call.

Once admitted, a call would then go to a packet sorter, where the packets are classified as real-time or non-real-time and placed in a packet queue. A call controller could then:

• Schedule different classes of packets
• Reduce bandwidth allocation resulting in graceful degradation
• Establish bandwidth reservation for high-priority applications
• Execute bandwidth compaction to maximize available channel resources
• Monitor radio resource usage.

Vigilance
In order to avoid the pitfalls that can develop during periods of intense network traffic, it's important to maintain ongoing surveillance of network performance. To accomplish this, a number of system tools are becoming available to support network-performance monitoring for a 3G network.

Signal-traffic management is one element of network management that gathers traffic data generated by various SS7 network elements, signal transfer points and service control points. These are important data items to compare against historical thresholds. When current thresholds begin to exceed their limits, alarms in the form of exception reports or e-mails can be generated for the network manager. Another element is network-traffic management, where network-traffic-management centers can optimize traffic flows during periods of network stress caused by traffic overloads. In this environment, traffic controllers through the application of various software tools (controls) are able to optimize the traffic-carrying capacity of the network.

Networks require an ongoing review of the carrying capacity. This is accomplished through the use of software tools that support the data collection and analysis of voice and signaling information to ensure that network resources have been properly distributed and that capacity has been accurately planned to support projected loads. The goal of this process is to ensure the maximum usage of network facilities to include network switches and supporting facilities.

Another element is traffic-data management where data for voice, data and signaling elements can be analyzed against stored thresholds to determine the adequacy of network resources. The traffic data is used to characterize customer-usage and -calling patterns to further identify changes in network resources. Coupled with this information is the packet-traffic-management function, which addresses the management of packetized data to allow service providers to manage their voice over IP (VoIP) and voice over ATM services.

What's Out There?
There are a number of systems being developed by Ericsson, Lucent, Motorola, Nortel and several independent software vendors such as Granite, Quintessent and Wisor. Some systems are aimed at supporting the wireless-service provider in a multivendor suite to support real-time network-traffic management, map network measurements into meaningful data and provide an end-to-end network overview. They also will support VoIP packets as well as establish monitoring and related control functions.

Lucent has developed as part of its Netminder system the Packet Traffic Management feature set. This system is designed to provide the service provider with the means to monitor and manage data-packet loads in near real-time. This allows the service provider to handle unexpected high- packet volumes across multiple service layers (voice and data) to optimize network resources. This system supports operating in near real time, giving the service provider ongoing packet network performance. This also allows the service provider to detect and manage network-performance events such as congestion before they can affect the end user.

Data collection is based upon periodic (5-minute) SNMP polling. Key data is collected for usage, packet loss and throughput for various network elements and individual ports. The system establishes normalization of performance data and logical data models to support data analysis and data display. There are provisions to set flexible threshold capabilities to facilitate alerts for network congestion and other network service parameters. Alerts can be sent via e-mail, pagers or fax.

There is support for a variety of network elements to include Cisco series (7500/12000), Ascend, Lucent, etc. The architecture is such that additional network elements can be added as they are implemented on the network.

Other features include near-real-time GUI updates for up to five simultaneous users, 2-D logical network-map display of performance data with icons to further access additional information for each element on the map. There is a query facility to access historical data that can be stored for up to eight consecutive days with 32 alerts. This system can be run on a standard Windows/NT server.

Something to Think About
Jeff Rinscheid, Lucent Technologies director of industry marketing, said that eventually service providers likely will bill for data services including QoS. He explained that service providers are content billing just for minutes used since wireless Internet access is an excellent strategy to build the "minutes of use" column. However, the pressure for QoS will begin to build when the providers begin to bill for content (Web data) and QoS.

With the rush to make 3G networks a reality and the headlong rush for greater Internet access, we all may wake one day and find that during a busy hour, our wireless handsets can't get dial tone. Early on, both Amazon.com and eBay were victims of a focused network overload. Until IP networks begin to develop the form of call control that we have come to appreciate with voice networks, they still are vulnerable. Although subscribers have become enamored with the "data service provider," we should not treat voice telephony with data as old hat.

Llana (llana@bellatlantic.net) is a Vermont Studies Group consultant.