In present work we address a problem of optimal resource partitioning in high-speed network with heterogeneous traffic classes and differentiated QoS requirements. The problem arguments are bandwidth and network-edge buffer space allocations for end-to-end connections. The problem formulation can take into account sensitivity variation of application performance for both aggregated and single service requests. The proposed method is recursive and solves two problems on each iteration step. The first step defines optimal relations between network media, equipment performance parameters and selected routing scheme in accordance with generalized mean of links and network nodes loads. Mutual influence of flows in the network is considered on this stage. The second stage is intended to provide optimal bandwidth and buffer size allocation parameters under network resource and QoS constraints. At this stage, stochastic resource allocation problem formulation with piecewise linearization of non-linear constraints is applied. We use connection request distribution, average connection duration time and quantity of information exchange as sensitive parameters in post optimal analysis to improve solution convergence to optimal tradeoff between resource allocation and acceptable application performance.
展开▼
