Item type |
Symposium(1) |
公開日 |
2017-10-04 |
タイトル |
|
|
タイトル |
Consideration on Applying Q-Learning to Backpressure Routing Algorithm to Improve Delay Performance |
タイトル |
|
|
言語 |
en |
|
タイトル |
Consideration on Applying Q-Learning to Backpressure Routing Algorithm to Improve Delay Performance |
言語 |
|
|
言語 |
eng |
資源タイプ |
|
|
資源タイプ識別子 |
http://purl.org/coar/resource_type/c_5794 |
|
資源タイプ |
conference paper |
著者所属 |
|
|
|
Nara Institute of Science and Technology |
著者所属 |
|
|
|
Nara Institute of Science and Technology |
著者所属(英) |
|
|
|
en |
|
|
Nara Institute of Science and Technology |
著者所属(英) |
|
|
|
en |
|
|
Nara Institute of Science and Technology |
著者名 |
Juntao, Gao
Mnoru, Ito
|
著者名(英) |
Juntao, Gao
Minoru, Ito
|
論文抄録 |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Back-pressure routing algorithm becomes increasingly popular for queueing networks, such as wireless ad hoc networks and MANETs. However, it is well known that backpressure routing algorithm has poor delay performance under light and moderate traffic loads. Available works propose to exploit shortest path and global queue length information to direct packets to shorter routes to their destinations to reduce packet delay. However, shortest path based backpressure routing inclines to route packets to shortest paths regardless of traffic congestion, thus causing long packet delay as traffic load increases. Global queue length based backpressure routing algorithm requires perfect knowledge of global queue length information which is hard to collect in real queueing networks. In this paper, we propose a Q-learning based backpressure routing (QL-BP) algorithm, which estimates route congestion based on only local queue length information. Our algorithm cannot only dynamically direct packets to routes of less congestion and effectively reduce packet delay, but also retain all appealing features of backpressure routing : throughput-optimality, distributed implementation and low computational complexity. As verified by simulations, our QL-BP algorithm reduces average packet delay by more than 71% when compared to traditional BP algorithm under light and moderate traffic loads. By enhancing QL-BP algorithm with shortest path information, our QL-BP algorithm outperforms all other variants of backpressure routing algorithms. |
論文抄録(英) |
|
|
内容記述タイプ |
Other |
|
内容記述 |
Back-pressure routing algorithm becomes increasingly popular for queueing networks, such as wireless ad hoc networks and MANETs. However, it is well known that backpressure routing algorithm has poor delay performance under light and moderate traffic loads. Available works propose to exploit shortest path and global queue length information to direct packets to shorter routes to their destinations to reduce packet delay. However, shortest path based backpressure routing inclines to route packets to shortest paths regardless of traffic congestion, thus causing long packet delay as traffic load increases. Global queue length based backpressure routing algorithm requires perfect knowledge of global queue length information which is hard to collect in real queueing networks. In this paper, we propose a Q-learning based backpressure routing (QL-BP) algorithm, which estimates route congestion based on only local queue length information. Our algorithm cannot only dynamically direct packets to routes of less congestion and effectively reduce packet delay, but also retain all appealing features of backpressure routing : throughput-optimality, distributed implementation and low computational complexity. As verified by simulations, our QL-BP algorithm reduces average packet delay by more than 71% when compared to traditional BP algorithm under light and moderate traffic loads. By enhancing QL-BP algorithm with shortest path information, our QL-BP algorithm outperforms all other variants of backpressure routing algorithms. |
書誌情報 |
第25回マルチメディア通信と分散処理ワークショップ論文集
巻 2017,
p. 190-192,
発行日 2017-10-04
|
出版者 |
|
|
言語 |
ja |
|
出版者 |
情報処理学会 |