 |
The Sixth International Conference on Advances in Semantic Processing SEMAPRO 2012 September 23-28, 2012 - Barcelona, Spain |
| Tutorials |
T1. Call-level Performance Analysis of Wired and Wireless Networks
Ioannis D. Moscholios and Michael D. Logothetis
Call-level performance analysis of wired and wireless (3G) networks has become challenging again because of the elastic nature of emerging multirate service-classes. Multi-rate teletraffic loss models aim at assessing the call-level QoS of IP based networks with resource reservation capabilities but also for 3G wireless networks (e.g. UMTS). This assessment is important for the bandwidth allocation among service-classes (guaranteeing QoS), the avoidance of too costly over-dimensioning of the network and the prevention, through traffic engineering mechanisms, of excessive throughput degradation. Despite of its importance, the efficient call-level performance modelling and QoS assessment is a challenge in the highly heterogeneous environment of modern telecom networks due to the presence of complicated call arrival process. Efficiency in call-level QoS assessment requires recursive formulas for the calculation of the various performance indexes including Call Blocking Probabilities (CBP), utilization of resources and mean number of calls of each service-class in the system.
The teletraffic loss models presented in this tutorial either of wired or wireless networks have as a springboard the classical Kaufman-Roberts recursive formula, used to accurately determine the link occupancy distribution and consequently CBP in the so called Erlang Multirate Loss Model (EMLM). In the EMLM, calls of different service-classes have fixed bandwidth requirements, arrive to a link of certain bandwidth capacity according to a Poisson process, and compete for the available link bandwidth under the complete sharing policy (a new call is accepted in the link if its required bandwidth is available - otherwise the call is blocked and lost). The recursive calculation of the link occupancy distribution leads to the efficient determination of the various performance indexes in the EMLM.
After having explained the EMLM, we proceed to its extensions for wired networks by distinguishing the call arrival process into three categories:
i) Poisson process (random traffic, infinite number of traffic sources),
ii) Quasi-random process (smoother traffic than Poisson, finite number of traffic sources), and
iii) Batch Poisson process (more peaked and bursty traffic than Poisson where calls arrive in batches, while batches follow a Poisson process).
For each category, we consider three types of service-classes:
a) Stream service-classes, whereby arriving calls have fixed bandwidth requirements and fixed bandwidth while in-service (stream traffic),
b) Elastic service-classes, whereby arriving calls have fixed bandwidth requirements but may tolerate bandwidth compression/expansion while in-service with a corresponding increase/decrease in their service time (elastic traffic), and
c) Adaptive service-classes, whereby arriving calls have fixed bandwidth requirements but may tolerate bandwidth compression/expansion while in-service without a change in their service time (adaptive traffic).
Furthermore, the notion of retrials for blocked calls is considered in the case of Poisson process and elastic/adaptive service-classes. To guarantee a certain QoS for each service-class in a multi-service loss system, we apply the Bandwidth Reservation (BR) policy in the aforementioned models.
For wireless networks, we firstly provide details on modelling the wireless environment of Code Division Multiple Access Systems, and then present EMLM-based models by distinguishing the call arrival process into two categories: i) Poisson process and ii) Quasi-random process. For each category, we consider stream service-classes, while we consider not only originated (new) traffic but also handoff traffic. Furthermore, we present models where calls may arrive to the system with several contingency resource requirements, while their request is made according to thresholds, which indicate the total number of occupied resources.
