Document Type

Journal Article


This paper proposes TrainNet, a vehicular network that uses trains to transport latency insensitive data. TrainNet augments a railway network by equipping stations and trains with mass storage devices; e.g ., a rack of portable hard disks. TrainNet has two applications. First. it provides a low cost. very high bandwidth link that can be used to deliver non real-time data. In particular. network operators can use TrainNet to meet the high bandwidth requirement associated with Video on Demand (VoD) services. Moreover. TrainNet is able to meet this requirement easily because its links are scalable. meaning their capacity can be increased inexpensively due to the continual fall of hard disk price. Secondly. TrainNet provides an alternative. economically viable, broadband solution to any regions that are reachable via a railway. Therefore, using TrainNet, any communities closed to a train station will be able to gain access to bandwidth intensive digital contents such as music, video, television programs, and movies cheaply. A key problem in TrainNet is resource scheduling. This problem arises because hard disks have finite capacity, and only a fixed number of hard disks can be loaded and unloaded at each station. To this end, this paper proposes three max-min scheduling algorithms, namely local max-min fair (LMMF), global max-min fair (GMMF) and weighted global max-min fair (WGMMF), to ensure the space on hard disks are divided fairly amongst competing stations. To study these algorithms, a simulator is constructed using the DESMO-J framework to investigate the behavior of these max-min schedulers in scenarios with realistic traffic patterns. Results show that while LMMF is the fairest algorithm, it results in data loss and has the longest mean delay, the lowest average throughput, and the lowest hard disk utilization. Further more, according to Jain's fairness index, WGMMF is the least fair algorithm. Despite that, it avoids data loss as is the case with GMMF. and achieves the best performance in terms of mean delay, averaged throughput, and hard disk utilization.

All rights reserved