On the Network Throughput Under Delay and Burstiness Constraints
Abstract
The rapid growth of delay sensitive applications, such as VoIP1 and IP video, requires the performance analysis of systems with quality of service (QoS) support. A fundamental question in this regard is to predict the traffic carrying capacity of a communication network by taking into account the bursty nature and delay requirements of sources which we call the delay constrained throughput in this thesis. The apparently simple problem poses serious challenges due to lack of an efficient tool for the modeling of delay and burstiness. In this work, we make use of network calculus (NetCal) as a tool to model delay and traffic burstiness. In addition, we also obtain the delay performance limit of communication networks.
We propose a method to quantify the delay constrained throughput of a wireline network having arbitrary topology which implements aggregate scheduling. A crucial intermediate step for throughput calculation is the derivation of the delay bound which is the maximum delay a packet may experience in the network. To this end, we make use of guaranteed rate analysis and deterministic NetCal. The proposed method is quite general and can be used for a variety of traffic specifications. It highlights among others, the effect of burstiness, delay guarantee and network diameter on throughput.
We take our methodology one step forward by predicting the delay constrained throughput of wireless networks. In this thesis the focus is on single-input singleoutput, multiple-input multiple-output and code-division multiple-access wireless networks. We can only provide stochastic service guarantees for wireless networks and thus we make use of stochastic NetCal to model the delay and burstiness. More specifically we make use of moment generating functions (MGF) based stochastic NetCal.
For the single-input single-output (SISO) system, we consider the ON/OFF source as the traffic model, while a two state Gilbert-Elliot (GE) model is used for channel modeling. We carry out the performance analysis of SISO channels where we study the impact of source burstiness, channel memory and signal-to-noise ratio (SNR) on delay constrained throughput.
For multiple-input multiple-output (MIMO) channel, we use the GE model to describe individual spatial paths between the antenna pairs. The overall channel is modeled by a J-State Markov chain, where J depends upon the degrees of freedom (DOF) available of the MIMO channel. We prove analytically and then show numerically that DOF based modeling is indeed accurate in the high SNR regime. Our proposed technique of DOF based Markov modeling results in a reduction of the state space from exponential in the number of antennas to linear in the DOF. The methodology is applied to the correlated MIMO model using virtual channel representation and then to the finite scatterers MIMO channel model. For virtual MIMO channel model, we study the impact of the delay requirements, violation probability, and the number of antennas on the delay constrained throughput under varying fading speeds and signal strength. Finally for the finite scatterers MIMO channel model, we study the impact of an increase in the number of antennas, channel burstiness, signal strength and fading speed on the delay bounds.
The method of quantifying the throughput is then applied to a code-division multiple-access (CDMA) multiuser system. We make use of large system analysis to simplify the physical layer model in which both the number of active users and the spreading factor grow without bound. At the physical layer, we have linear minimum-mean square error receiver and adaptive modulation and coding while the channel service process is modeled using a finite-state Markov chain. We study the impact of delay requirements, violation probability, and the user load on the traffic carrying capacity under different signal strengths. A key insight provided by the numerical results is the extent one has to back-off from capacity under the different delay requirements.
Multiuser detectors (MUDs) are used to separate the signals of individual users in a non-orthogonal CDMA system. It is well-known that the choice of the MUD has a significant effect on the system capacity. We quantify the effect of linear MUDs on the traffic carrying capacity of uplink CDMA channels under delay and burstiness constraints. The following receivers are considered for performance evaluation: single-user matched filter, decorrelator, and linear minimum mean square error detector. We quantify the effect of these multiuser receivers on the traffic carrying capacity of uplink CDMA channels under different delay requirements, violation probability, and the user load.