Providing quality-of-service guarantees in multi-service wireless networks

Abstract

Providing quality of service (QoS) guarantees over wireless packet networks poses a host of technical challenges that are not present in wireline networks. One of the key issues is how to account for the characteristics of the time-varying wireless channel and for the impact of link-layer error control in the provisioning of packet-level QoS. In this dissertation, we accommodate both aspects in analyzing the packet loss and delay performance over a wireless link. We also propose novel techniques for quantifying the wireless effective bandwidth, defined as the minimum amount of bandwidth that needs to be allocated to ensure a given level of QoS. These techniques are essential to on-line connection admission control (CAC) and capacity dimensioning in multiservice wireless networks with QoS support. To analyze the loss and delay performance, we consider a wireless link whose capacity fluctuates according to a fluid version of Gilbert-Elliot channel model. Incoming traffic sources are modeled with on-off fluid processes, which capture the bursty nature of network traffic. The packet loss performance is analyzed for the cases of a single and multiplexed traffic streams. For the single-stream case, we derive the packet loss rate (PLR) due to buffer overflow at the sender side of the wireless link. We also obtain a closed-form approximation for the corresponding wireless effective bandwidth. In the case of multiplexed streams, we obtain a good approximation for the PLR using the Chernoff-Dominant Eigenvalue (CDE) approach. The delay performance is analyzed via two distinct yet complementary approaches: fluid queueing analysis and discrete-time analysis, each being advantageous in analytic tractability and accuracy, respectively. The fluid approach is used to derive the packet delay distribution via two different approaches: uniformization and Laplace transform. Using the analytic results, we investigate the packet discard rate at the receiver, which is particularly important for delay-sensitive traffic. The delay distribution is further used to quantify the wireless effective bandwidth under a given delay guarantee. Numerical results and simulations are used to verify the adequacy of our analysis and to study the impact of error control on the allocation of bandwidth for guaranteed packet loss and delay performance. Finally, we use discrete-time analysis to quantify the mean delay experienced by a Markovian source over a wireless channel. In this case, the wireless link implements the selective-repeat automatic-repeat-request (SR ARQ) scheme for retransmission of erroneous packets. We obtain good approximations of the total mean delay, which consists of transport and resequencing delays. The transport delay, in turn, consists of queueing and transmission delays. The exact probability generating function (PGF) of the queue length under "ideal" SR ARQ is obtained and combined with the retransmission delay to obtain the mean transport delay. For the resequencing delay, the analysis is performed under the assumptions of heavy traffic and small window sizes (relative to the channel sojourn times). We show that ignoring the autocorrelations between packet interarrival times or the time-varying nature of the channel state can lead to significant underestimation of the delay performance, particularly at high channel error rates.