6.3 Performance AnalysisTo make the analysis tractable, we consider a simplified case that (1) there isone real-time mini-slot, and all the real-time packets are treated equally; (2)per-router fairness is considered. We assume that the voice and video call arrivals at each source node are independent and follow a Poisson process, and the call duration has an exponential distribution.6.3.1 Real-Time Traffic Access Delay BoundThe access delay is defined as the time period from the instant that a packet becomes the head in the buffer to the instant that the packet departs from the router. Let Ts denote the time duration of one slot, and Nm the number of mini-slots, including the real-time mini-slot. Consider the worst case that the target router has Nm 2 onehop and two hop neighbors, and all of them have real time packets to transmit. After the target router transmits one packet, it takes Ts (Nm 1)for the target router to transmit the next one. Thus, the access delay bound ofreal time traffic at each hop is Ts (Nm 1), which is independent to the traffic load of the networks.6.3.2 Data Traffic Access DelaySince guaranteed priority access is provided to real time traffic, the real timetraffic load will impact the data traffic access delay. The voice call is represented by an on/off model with parameters nd . At n on stte, voice pckets re generted periodiclly with n inter-rrivl time Io, while no voice pcket isgenerted t n off stte. For video cll, the video frmes re generted periodiclly with n inter-rrivl time Iv. The video frme usully hs lrge ndvrile size [93]. Suppose tht it tkes one slot to trnsmit one voice pcket, nd Mv slots (on verge) to trnsmit one video frme. Considering trget router, we refer to its two-hop vicinity s the trget re.To otin the dt trffic ccess dely, we first need to derive the frction ofchnnel time occupied y voice nd video trffic. We define two-dimensionl stte (nv, no), where nv nd no re the numers of video clls nd voice clls, respectively, eing served y the routers within the trget re. Denote the verge rrivl rtes of voice nd video clls tht trverse the trget re s o andv, respectivey, and the average ca duration as o1 and v1, respectivey. We assumethat ca admission contro is in pace to guarantee the QoS of voice and videocas, and the maxima number of acceptabe voice and video cas within the target area are denoted by No and Nv, respectivey. The state transition diagram is shown in Fig. 6.3. Since a video ca requires more resources than a voice ca, when there is 1 video ca being served, the maxima number of supported voice cas is No Mv, denoted by No m. Define pi j as the joint probabiity that i video cas and j voice cas being served.