The main goal of quality of service (QoS) is to help reduce or eliminate delay of voice
packets including packet loss that travels across a network. It can be defined as the
capability of a network to provide better service to selected network traffic over various
underlying technologies such as Frame Relay, ATM, and IP. This network feature helps
in differentiating different classes of traffic and treats them differently. The various
formal measurements of QoS are:
Measurement Description
Service Availability The availability of the users’ network connection and depends on the connected network device.
Throughput The packet delivery rate at the endpoints.
Delay The end- to- end packet delay, while traversing the network.
Delay jitter The delay variation among similar packets traversing the same path in the network.
Packet loss rate The rate of packet loss, because of packet dropping and corruption.
QoS of networks should attempt to maximize service availability and throughput, while at the same time minimizing the remaining measurements.
Need for QoS
In the current network scenario, different types of traffic (such as real-time and data) need to share the same network link. These different kinds of traffic require different treatment from the network. This is similar to a first class passenger of an airplane demanding preferential treatment over other passengers. Just as, apart from providing special treatment, a separate airplane cannot be made available to the first class passenger, similarly a separate link or a network connection cannot be given to different customers, though the treatment
given can be different. The entire bandwidth has to be shared between priority traffic and regular traffic, and only at places where the traffic flows through active network elements like routers, can these flows be differentiated and treated differently.
The different types of traffic can be grouped as:
! Non Real-time or Data: These applications only care about reliable packet delivery like that guaranteed in TCP. They are immune to delays or bandwidth requirements. Examples are web browsing, email, distributed computing, etc.
! Real-time: These applications require timely delivery along with reliability. Some of these applications can tolerate an upper bound in delay, whereas others are totally intolerant. This range of expectations is used to further classify real-time applications by QoS models.
QoS Service Models
There are three major architectures or service models for implementing QoS in packet
networks. These are:
QoS Service Model Description
Best Effort Service In this model there are no QoS guarantees given to the application, except that a best effort to deliver will be made. Integrated Service It is also called IntServ. This is an earlier model (1995) devised for
integrated service networks such as ISDN or ATM. It believed to have end-to-end QoS strategies implemented at the network elements for all classes of traffic flows.
Best Effort Service
This model allows the application to send any amount of data at will and without any
authorization. The network elements in turn will try their best to deliver the packets to their
destinations without any constraints of maximum delay, latency or jitter, etc. They also give
up trying to deliver after attempting for a number of times, in case acknowledgments from
the receiver are not received. The network will also not inform the sender that the attempt to
deliver has been abandoned. Thus, the end points have to take care and incorporate
reliability features within themselves. The IP network is an example of a Best Effort service
model.
Integrated Service
The Integrated service model is defined by a set of standards laid down by IETF. This model assures different QoS-profile treatment, dictated by the network elements, to the multiple classes of traffic. In this model, the applications are aware of the traffic characteristics that they would put on the network and accordingly signal the network elements to reserve required resources before sending their data. The network elements in turn acknowledge the signal positively if they are able to reserve the resources or else send a negative
acknowledgement.
This model categorizes applications, in terms of their network traffic requirements, into three classes. Real-time Tolerant (RTT), where some delays can be tolerated within a small range as in video or audio playback applications. Real-time Intolerant (RTI), which requires minimal or absolutely no delays, as in video conferencing. And finally, Elastic applications, where as long as the packets are delivered reliably no delay constraints are imposed, as in web browsing or email. Accordingly, the model provides for the following classes of service:
Guaranteed service: This service guarantees bandwidth and provides a deterministic upper bound on delay. It is used for RTI applications. Controlled Load service: This service guarantees an average delay, but the specific
end-to-end delay by some arbitrary packet cannot be specifically determined. The Integrated service network implements the following mechanisms to guarantee QoS levels.
Mechanism Description
Admission Control This allows the network (network elements) to refuse a new traffic flow request from an application, depending upon resource availability. This is usually a policy-based decision taken by the
router. Policy Enforcement Points (PEP) and Policy Decision Points (PDP) are components of this mechanism. PEPs and PDPs use a simple request response protocol called Common Open Policy Service (COPS) to communicate between themselves.