Deadline-Constrained Multi-Agent Collaborative Transmission for Delay-Sensitive Applications

Delay-sensitive applications impose the strict deadline-constrained latency requirement for end-to-end transmission. However, due to the high dynamics of link status and the limited network resources, it is still challenging to ensure deterministic end-to-end latency for deadline-constrained flows....

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on cognitive communications and networking 2023-10, Vol.9 (5), p.1-1
Main Authors: Liu, Kang, Quan, Wei, Cheng, Nan, Zhang, Xue, Guo, Liang, Gao, Deyun, Zhang, Hongke
Format: Article
Language:English
Subjects:
Active control
Algorithms
Artificial neural networks
Collaboration
Deadline-constrained transmission
Delay
Delay-sensitive applications
Delays
Heuristic algorithms
Job shop scheduling
Machine learning
Minimization
Multi-agent collaborative
Multiagent systems
Optimization
Queueing
Queues
Routing
Shortest-path problems
Stochastic optimization
Training
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Delay-sensitive applications impose the strict deadline-constrained latency requirement for end-to-end transmission. However, due to the high dynamics of link status and the limited network resources, it is still challenging to ensure deterministic end-to-end latency for deadline-constrained flows. To this end, we propose a deadline-constrained multi-agent collaborative transmission mechanism (DMCT). Specially, DMCT is featured by combining dynamic routing and packet active dropping control to avoid additional end-to-end latency. Firstly, we design a specific packet structure to detect network status and aware the deadline of each flow. Secondly, to enhance the precision of routing decision, we further formulate deadline-constrained flow scheduling as a joint minimization problem that considers the deadline of flows, queuing delay, and network status. Then, we propose a multi-agent collaborative routing algorithm to find the solution to the joint minimization problem. Finally, we propose a local-agent deadline-constrained active queue management algorithm for packet active dropping control and routing adjustment. Simulation results show that the proposed DMCT outperforms the state-of-the-art solutions, e.g., graph convolutional network-based deep reinforcement learning (GCNL) and delay-based shortest path first (D-SPF), in terms of one-way round-trip time, drop rate, and queue length.
ISSN:2332-7731
2332-7731
DOI:10.1109/TCCN.2023.3288133