| 摘要: | In this paper, we focus on the problem of translating a global QOS (quality of service) requirement into a set of local QOS requirement. Applications with guaranteed service would like to acquire an end-to-end delay bound from the network.
In the literature, most researches only attempt to solve the worse case delay bound suffered by a connection passed through a series of switching nodes. Simply, they equally allocate the excess delay or reserve the same bandwidth in each switch along the path. However, if the network can apply a tighter delay bound than the applications want, and the excess delay amount can be adequately distributed into the local switches along the path, applications can get the best possible performance.. It motivates us to propose a novel local QOS requirement allocation scheme to get better performance which is abbreviated by the supportable connection number. We proposed an analysis model to evaluate the allocation scheme in a series of switching nodes with the rate-controlled scheduling architecture, including a traffic shaper and a nonpreemptive earliest-deadline-first scheduler. With the number of maximum allowable connections as the performance index, we derived an optimal delay allocation policy. The numerical results have shown the importance of allocation policy and the factors that affect the performance index. |
