Scheduling strategies to improve quality of service for heterogeneous data over resource constrained wireless mesh networks

[摘要] ENGLISH ABSTRACT: Cost and bandwidth plays a major role in many telemetry and Internet of Things (IoT)application network implementations. Wireless Mesh Networks (WMNs) based on singleradiosingle-channel (SRSC) networks will likely attract more deployments if the requiredquality of service (QoS) can be provided. Carrier sense multiple access with collisionavoidance (CSMA/CA) is more widely used in WMNs where access to the network isdecentralized and each node makes its own decision on when to access the channel. CSMA/CA in WMNs faces an increase in packet loss and contention with an increase inthe number of hops to reach the destination, compared to nodes that are closer to thedestination, resulting in an increase in collisions and wastage of bandwidth as the packetshave to be re-transmitted. The distributed coordination function (DCF) approach doesnot provide data differentiated priority services, while the enhanced distributed channelaccess (EDCA) method was mainly designed for delay sensitive non-elastic applications toprovide differentiated services. With EDCA, an unfairness problem exists, where higherpriority data can starve low priority data under high load scenarios while EDCA performspoorly in multi-hop networks.This dissertation focuses on improving QoS by reducing packet loss, reducing collisionsand improving fairness to prevent starvation in low cost SRSC WMNs. This researchhypothesizes that these problems can be addressed by first selecting a packet for transmissionand then performing the channel contention by removing the internal contentionmechanism. It also asserts that the queue selection mechanism plays a critical role inthe achievable QoS. The research also hypothesizes that hybrid configured network layoutsusing DCF can improve performance. Hybrid and homogeneous configured networklayout strategies have been investigated to support the research. Five medium accesscontrol (MAC) layer schedule before contention (SBC) mechanisms have been developed,namely adaptive weighted round robin (AWRR), roulette wheel sampling (RWS), RWSAGE,congestion control and fairness scheduling strategies (CCFS) and queue load control priority (QLCP). The performance of these strategies is compared to EDCA and DCFnetworks through simulations.It was found that the CCFS mechanism tends to starve lower priority data underheavy loads and performs poorly. RWS-AGE showed the least packet loss in homogeneousconfigured network layouts. A random weighted selection strategy with an age counterperforms better than a weighted round robin strategy. If the lower priority data are notstarved, it helps to lower packet loss as they use larger CW ranges for the back-off. Animportant ingredient in reducing packet loss in hybrid con gured network layouts is byusing DCF as it has a larger CW range and also reduces collisions. To further verify thatRWS-AGE reduces packet loss, the strategy was implemented on the FIT-IoT Lab testbed and its performance was veri ed. A novel analytical model for the end-to-end delayfor SBC strategies following Markovian theory has been developed. The results fromthis research support the initial hypotheses and provide important guidelines for networkimplementation in resource constrained WMNs carrying heterogeneous elastic traffic fora variety of applications.

[发布机构] Stellenbosch University