 The QS performance parameters, which are supposed to be the performance metrics, which are measurable actually determine how the service is going to be provisioned from a service provider to the end user. So we are going to discuss it from the protocol's perspective. Then we are actually going to look at the performance parameters and see how different IP traffic classes are influenced or affected by these performance parameters. The protocols being central to the implementation of the NJN architecture for service provisioning are very important. When we are talking about IP based network services and in all IP environment, we are talking about the protocols which work with IP above and below. So at the lower layer, we are expecting some kind of connectionless and connection oriented services like Ethernet, ATM, SDH, PDH, Wi-Fi, that is 802.11 ETC. At the higher layer, we are expecting the application layer services like RTP, RTSP, RTCP, session initiation protocol and at the transport layer, we have TCP, UDP or sometimes neither of them. So if that is the case, we can say that these protocols are going to have a final role in determining how the QS is actually provided. If you look at this diagram, we have an all IP environment. We have end to end scenario where we have the service provider and we have the service seeker in a source destination situation. You can see that at the higher layer, there are so many protocols which affect which use IP and which affect the behavior of IP. Likewise, the data link layer protocols also determine how this IP packet, which is an hourglass model is affected in terms of service provisioning. So from the pure IP perspective, there are certain parameters which define exactly how a QS class can be provisioned services. The first one is the IP packet transfer delay. It is actually the difference in the occurrence of two transmission events because one transmission event is followed by the second transmission activity. So this packet transfer is a measure that can be given any form. For instance, it's an average transfer delay, the max transfer delay, minimum transfer delay or the median transfer delay. So this is very important because from the network perspective, each intermediate device such as a router or a switch is processing an IP packet. Once this delay is there, then any variation in it might cause jitter. So this is actually the difference between one way delay of IP packet and it is compared to the reference IP packet transfer delay, which can vary of course, but there is some kind of guideline on how to choose the reference IPTD. So whenever this IPTD is deviated, then depending upon the standard deviation or variance, this IP packet delay variation can be computed. It's another very important parameter. Then we have the packet error ratio. It is the total number of packets which are either corrupted or lost because if you recall in the IP header, there is checksum. That determines the erroneous or corrupted packet which is discarded by a router or an end host. Then we have the packet loss ratio. That is a packet that is processed by a router sent to another router but is never received or it is dropped at the input queue of a router buffer. So this is actually another interesting perspective on the total capability of a network to deliver a certain number of packets. This in turn requires some additional information on the overall capability of the network in terms of bandwidth, in terms of the buffer available to every router, etc. So the IP traffic that is carried on could be, we are talking about QS requiring IP traffic. So there are QS classes which are based on the IPTD, IPDV and error and loss ratios. So you see we have different classes. For these eight classes, in fact class five is not specified and class six and seven are quite ambitious. So they are not offered by different service providers but for the first few classes that is from class zero to class one, we can look at them from their usability point of view. So class zero and one are for real time delay and jitter sensitive applications. So most jitter sensitive applications are sent on to class zero and one, specifically class zero being the best. It is highly suited for high interaction applications such as voice over IP because it has very low bound on IPTD. Class two and three are targeted for transactional data again interactive but having relatively more IPTD. For instance, class two suits the signaling traffic and class three is for interactive applications. Now class four relatively has more IPTD. So it is for short transactions which are again live but not requiring much jitter sensitivity. And video streaming is another interesting application of class four IP traffic class.