Quality of Service in Next Generation Networks

1 NGN Introduction
The Next Generation Network (NGN) is to provide various multimedia communication services under a common network infrastructure, across different service providers and without spatial limitations. The NGN initiative is championed by the International Telecommunication Union (ITU) and is supported by almost all the major service providers and equipment manufacturers.
    Once upon a time, the mission statement of NGN was shared by the network based on the ATM technology. The "various communication services" part of the NGN objective can be satisfied by the ATM based network as long as the services are connection-oriented; and the "without spatial limitation" part of the NGN objective can only be fulfilled as long as it is not mobile and having a fixed and peculiar ATM address. However, for the two pillars of today’s communication services (i.e., connectionless IP services and mobile services), the ATM based network missed both. Besides, the ATM friendly voice network becomes a hurdle to the "promised land" and is the starting point of the "network migration".

    With the current broadband penetration and mobile IP, it can also be argued that the NGN promise has been at least partially realized by today’s IP network. All the service providers have to do are to increase the network capacity and to broaden the network reachability. There are certain merits in this argument; the service providers are providing various mobile and non-mobile audio/video (i.e., Multimedia) services, only the quality part is absent.
One of the key objectives of the NGN initiative, among numerous other objectives regarding the network transport and service capabilities, is to provide Quality of Service (QoS) for the multimedia content. In this article, various aspects of QoS will be discussed, with their implications on the network architecture.

2 Decomposition of QoS
QoS can be a highly technical term. Under different context, its meaning may be different. For example, ITU-T Recommendation E.800 defines QoS as "the collective effect of service performance which determines the degree of satisfaction of a user of the service." As evident in this definition, the QoS will be a highly subjective and application sensitive measure and such measure can hardly be constructed as a measurable and objective QoS definition to measure the network performance.

    For a packet-based network, the QoS is normally measured by the parameters like time delay, delay variation and packet loss ratio (e.g., ITU-T Y.1541 on IP QoS). Those parameters can then be further refined to route selection, the queuing priority and the discard priority on each network element along the path from the content source to its destination.

    The merit of such parametric QoS definition is measurable and, for given traffic source characteristics, it is even possible to allocate network resources, which ensures an upper bound on those above-mentioned QoS parameters.
This type of parametric QoS definition regarding the network performance may be more applicable to the packet transport services between two different business entities than to a typical user of the network.

    For a typical user of the network, such parametric definition of QoS provided by the network would be of little concern. A network, as perceived by its user, needs to be "good enough". For any application, there exists a range of QoS parameters mentioned above, which gives application performance levels ranging from "barely acceptable" to "excellent".

    An astute reader may notice that a bridge is required to link the application level performance to the network performance, characterized by the parametric QoS parameters (i.e., delay, jitter, loss probability, etc). Since it is impractical to assume that the network is aware of applicability, the burden of such bridging naturally becomes the responsibility of the applications.

    If dedicated network resources are to be reserved for certain application, since the source of the traffic does not aware, or not likely to aware, the impairment imposed by the network along the path, a traffic source can only provide description of the traffic to be generated. On the other hand, the receiver could have the information on both the traffic characteristics from the source and the accumulated impairments from the network. Logically, the receiver should be the one to request network resources, based on the required performance level from its user, from the traffic information from the source, and from the possible impairments from the network. This is also one of the design objectives of RSVP, a resource reservation protocols for the IP based networks.

    In summary, QoS concerns the parametric network performance parameters and the application
"satisfaction" level. The receiver is in the best position to select the service parameters (i.e., requesting relevant network resources) since the network impairments are unknown at the source of the traffic.

3 Different QoS Approaches
There are different network architectures to support the content transport with QoS. The most common approaches can be outlined as follows:

• Over-Provisioning: normally over-provisioning applies only to the portion of the network where bandwidth is abundant. If bandwidth is available, such an approach requires very little, if any, network resource control. There is almost no per-packet based processing with the exception of queuing priority.
• Over-Provisioning with Traffic Engineering: This approach is an enhancement to the over-provisioning approach. If the network size is reasonably large, the network may be severely under-utilized without traffic engineering. Traffic engineering normally involves distributing the traffic evenly across the network so that there is no particular bottleneck links inside the network. The per-packet processing may involve route selection for the same destination.
• Differentiated Services: Differentiated services approach allows the network to insert more control on the per-packet granularity. The per-packet processing encompasses route selection, queuing priority assignment and the discard priority assignment. The network will also need to dedicate resources to a particular traffic class.
• Per-Flow Reservation: The network resource reservation is per-flow based. A flow can be either a fine granular flow (e.g., IP flow identified by source address, destination address, source port, destination port and protocol ID) or an aggregated flow, where multiple fine granular flows combined. This type of network is normally resource constraint.
  

  

    From the view of control complexity, the Over-Provisioning approach requires minimum resource coordination. Hence it will be the most scalable approach but it would also be hard to ensure parametric QoS under all the possible scenarios without drastically under-utilize the network capacity.

    In order to improve the network utilization, traffic engineering techniques, with the advent of the MPLS technology, has been introduced. This will alleviate the bandwidth utilization issues in the network but cannot eliminate the fundamental issue related to QoS, which is the dynamic resource allocation based on need.

    Differentiated Service is introduced to further enhance the network utilization and to allocate resources properly among the competing parties. Normally, differentiated service techniques will be used in the core of the network where micro-management of the resources is impractical, if not impossible. It should also be noted that the traffic engineering techniques also be used in the differentiated service environment to further enhance the network resource utilization. Because of the lack of resource dedication to a specific traffic flow, the parametric QoS (defined as delay, jitter and loss probability) cannot be ensured by a network where Differentiated Service techniques are used as the sole QoS control mechanism.

    In order to provide parametric QoS, the per-flow based resource reservation is required. In the per-flow based resource reservation setup, the "state" of the flow, which can be a micro-flow or an aggregated flow, will need to be persistent inside the network during the duration of the flow. Hence the number of flow supported by the network would be limited.

    IntServ model is a typical approach for per-flow reservation. To enhance the scalability, bandwidth broker (BB) is introduced to avoid keeping the flow status along the data path. And the BB could do the admission control and resource allocation for the whole network. Usually BB works together with differentiated service network.
In a network, a combination of the above QoS approaches could be employed, depending on the nature of the network and the nature of services provided.

4 Controlling Aspects of QoS
As mentioned above, the overall network may actually consist of multiple QoS networking architectures. Depending on different networking QoS architectures, the control mechanisms and control complexity will be different. The resource controller in the control plane is dedicated to perform the control functions related to QoS assurance. The resource controller may or may not co-exist with the data plane and the resource controller may control one or multiple network elements on the data plane.

    Regardless of the QoS architecture employed, the resource controller may command the data plane to impose limitations for traffic ingress to the network. The purpose of such limitations is to ensure the stability of the network and also to honor Service Level Agreement (SLA).

    In the per-flow reservation based QoS architecture, signaling protocols may be used between the traffic source and the traffic sink in order to secure the required resource. It is important to note that those signaling protocols need to tunnel through the portion of the network where QoS guarantee is not based on per-flow reservation.

    For the portion of the network where QoS guarantee is based on the over-provisioning architecture, the resource controller in the control plane may also participate in the resource management for traffic engineering purposes. The functions performed by the resource controller in managing resources may involve real time resource usage monitoring at various locations inside the network. The results of such monitoring may lead to dynamic or static traffic flow realignment.

    In the differentiated service based QoS architecture, for traffic engineering purposes, the resource controller needs to execute all the functions as in the over-provisioning based network. In addition, the resource controller may also be responsible for advertising its capabilities via proper protocols. Those advertisements will be useful if the network is crossing administrative boundaries.

    For the network using per-flow reservation based QoS architecture, the major challenge is scalability. Even though such QoS architecture provides fine granular resource control and the possibility to provide parametric QoS guarantee, there are still questions on:

• if the benefits of such resource control warrant the control complexity.
• if it is practical in a large scale network.
  In today’s network, RSVP (or RSVP-TE) is the most popular protocol for such purposes. Even with many iterations of refinements, scalability still remains to be the top issue associated with the per-flow reservation based QoS architecture.
  

5 Business Model
Services with QoS have been studied for many years, but until now hard QoS guarantee is still not deployed in the current IP network on a large scale. Lack of a clear business model hinders the advent of QoS in providing IP services.

    Best effort services and flat-rate billing have been the norms in providing IP services. Operators begin to realize that broadband access services can only produce limited new revenue when the penetration of broadband is almost saturated. Value-added services need to be devised for generating a new revenue stream. Services with QoS assurance has been viewed as such services. Different grades of services could be charged at different price levels.
However, the accounting and billing policy for services with QoS has not been clearly defined yet. Traditional
fine-granular accounting and billing practice used in the telecommunication industry may lead to a high complexity in network equipment and operations, if such a scheme is feasible at all due to the nature of the IP packet transport
(e.g., how to count the retransmission). Scalability is yet another obstacle facing the network operators as the network scales. The cost of the accounting and billing operation is also another major issue for the operators.

    In current networks, free voice and video services can be obtained and in most cases the quality of service is acceptable, which means that the network is "good enough". In this case, the demand to enhance the network QoS, where charge is to be applied, is unconvincing. It is still uncertain how many people will seek premium services with QoS if the best-effort service is acceptable.

6 Conclusion
QoS has become one of the most important issues in the next generation network. Different from the original research on the single node mechanism, current focus of QoS study has shifted to the end-to-end QoS control, such as QoS signaling and bandwidth broker. Due to lack of a good business model and clear accounting and billing architecture, QoS is still under way.

References
[1] ITU-T Recommendation Y.1540. Internet Protocol Data Communication Service —IP Packet Transfer and Availability Performance Parameters [S].
[2] ITU-T Recommendation Y.1541. Network Performance Objectives for IP-Based Services[S].
[3] ITU-T Recommendation Y.1291. An Architectural Framework for Support of Quality of Service (QoS) in Packet Networks[S].
[4] IETF RFC 3726. Requirements for Signalling Protocols[S].
[5] IETF RFC 3260. New Terminology and Clarifications for DiffServ[S].

Manuscript received: 2004-10-08