Author:

Keywords:

Hybrid Simulation, Queueing Network, Wireless Telecommunications, Stochastic Simulation

Abstract:

Evaluating the performance characteristics of telecommunication networks is a challenging task. Measurements of relevant data on an actual running system are often impossible. Therefor, analytical models are proposed in literature. However, even after simplifying assumptions and decompositions, the resulting analytical model is usually mathematically intractable. The only alternative for predicting the performance of complex telecommunication networks is a simulation.A packet-switched communication mean can be modeled by a queueing model. In queueing theory, a queue consists of a buffer, i.e. a waiting area,and a service zone. The queue itself corresponds to a switching node ora router having both a packet buffer and an associated outgoing interface with a limited transmission capacity. A telecommunication network canbe represented by interconnecting the queues. An event simulator generally speaking mimics the behavior of the queueing network by tracing the customers, i.e. the packets, using events.In a simulation experiment, samples of many thousands of observations are often required to estimate some performance measure with sufficient confidence. In view of computer run time, it is clearly important that one should attempt to reduce the volume of sampling in order to reduce therun time of the simulation. Unfortunately, it has been a common practice to apply simulation techniques uncritically with little thought to theefficiency of simulation runs. Hybrid simulation is a way to reduce thesimulation run time without compromising the precision of the estimatesby combining analytical solutions to the simulation experiment with event-driven simulations having a reduced number of events. The splitting of the model in an analytical part and a simulation part is a crucial element of a hybrid simulation. Considering a packet base telecommunicationnetwork, the packetized traffic can be split in two parts: the foreground packets, directly related to the performance measures of interest, and the background stream, having only an indirect relation with the performance measures of interest. In the novel stochastic hybrid simulation scheme detailed in this thesis, the splitting of the traffic into foreground and background traffic is combined with the general idea of using a stochastic technique to evaluate the impact of the background traffic onthe foreground packets. At the queue level, a stochastic fluid-flow models the background packets and the queueing delay of the foreground packets is estimated by exploiting the ASTA property, i.e. when the lack of bias assumption is respected, the waiting time of a packet from a foreground source has the same distribution as the virtual waiting time of theaggregation of all traffic streams at the queueing system.In a second part of the thesis, the stochastic hybrid simulation is extended to the simulation of queueing networks of wireless nodes sharing acommon channel. As soon as shared channels are dealt within a packet-switching mode, conflicts arise when more than one packet is simultaneously transmitted using the same channel. Whenever a portion of one user's transmission overlaps with another user's transmission at the receiver's side, the two interfere and “destroy” each other. The problem to be solved is how to control the access to the channel in a way which produces, under the physical constraints of simplicity and hardware implementation, an acceptable level of performance. The difficulty in controlling a channel which must carry its own control information gives rise to the so-called random-access modes. The MAC protocol together with the transmission capacity on the channel and the distribution of the source-destination pairs determines the saturation throughput of the system transmitter-channel-receiver. In reality not every transmitter has always a packet ready to be transmitted and some nodes can hold more than one packet in apacket queue waiting to be transmitted. The notion of capacity region based on the saturation throughput is needed to extend the stochastic hybrid simulation scheme for wireless packet radio networks. When a foreground packet arrives at a queue during the simulation, not only a sampled value of the queueing delay for the packet in the queue is provided but also the associated instantaneous transmission capacity as seen by the node.The evaluation of the saturation throughput in a realistic multi-hop scenario is a non-trivial issue. Two semi-analytical models for the saturation throughput of existing wireless IEEE Std 802.11 and IEEE Std 802.15.4 standards are proposed. Both models are supported by detailed simulations and the latter is also validated on a testbed of wireless motes.