Notations and performance metrics

We denote by $\lambda$, $L$, and $T$ as the average arrival rate of clients, the length of the video, and the threshold value, respectively. We use the following performance metrics in our evaluation: (1) Rejection probability ($p$) - the probability that a client cannot be admitted and served; (2) Normalized workload ($W$) - the product of average client arrival rate and video length, $W = \lambda L$; (3) Effective normalized workload ($W_e)$ - the admitted normalized workload, $W_e = W (1-p)$; (4) Average workload placed on the network/Network usage ($U$) - the sum of all link usage, $U = \sum_i u_i$, where the link usage of link $i$, $u_i$, is the product of link utilization and its capacity. Hence network usage represents the average workload placed on a network; (5) Server stress ($S$) - the average amount of bandwidth used at the server; (6) Startup delay ($d$) - the time that clients have to wait before actually receiving the video and playing it back. In the following simulation we assume that the video length is 100 minutes.