TY - GEN
T1 - IFS-RL
T2 - 2018 ACM SIGCOMM Workshop on Network Meets AI and ML, NetAI 2018
AU - Zhang, Yi
AU - Xu, Kuai
AU - Bai, Bo
AU - Lei, Kai
N1 - Publisher Copyright:
© 2018 Association for Computing Machinery.
PY - 2018/8/7
Y1 - 2018/8/7
N2 - Named-Data Networking (NDN) is a new communication paradigm where network primitives are based on named-data rather than host identifiers. Compared with IP, NDN has a unique feature that forwarding plane enables each router to select the next forwarding hop independently without relying on routing. Therefore, forwarding strategies play a significant role for adaptive and efficient data transmission in NDN. Most of the existing forwarding strategies use fixed control rules based on simplified or inaccurate models of the deployment environment. As a result, existing schemes inevitably fail to achieve optimal performance across a broad set of network conditions and application demands. In this paper, We propose IFS-RL, an intelligent forwarding strategy based on reinforcement learning. IFS-RL trains a neural network model which chooses appropriate interfaces for the forwarding of Interest based on observations collected by routing node. Not relying on pre-programmed models, IFS-RL learns to make decisions solely through observations of the resulting performance of past decisions. Therefore, IFS-RL can implement intelligent forwrarding which adapt to a wide range of network conditions. Besides, we also researches the learning granularity and the enhancement for network topology change. We compare IFS-RL to state-of-the-art forwarding strategies in ndnSIM. Experimental results show that IFS-RL can achieve higher throughput and lower packet drop rates.
AB - Named-Data Networking (NDN) is a new communication paradigm where network primitives are based on named-data rather than host identifiers. Compared with IP, NDN has a unique feature that forwarding plane enables each router to select the next forwarding hop independently without relying on routing. Therefore, forwarding strategies play a significant role for adaptive and efficient data transmission in NDN. Most of the existing forwarding strategies use fixed control rules based on simplified or inaccurate models of the deployment environment. As a result, existing schemes inevitably fail to achieve optimal performance across a broad set of network conditions and application demands. In this paper, We propose IFS-RL, an intelligent forwarding strategy based on reinforcement learning. IFS-RL trains a neural network model which chooses appropriate interfaces for the forwarding of Interest based on observations collected by routing node. Not relying on pre-programmed models, IFS-RL learns to make decisions solely through observations of the resulting performance of past decisions. Therefore, IFS-RL can implement intelligent forwrarding which adapt to a wide range of network conditions. Besides, we also researches the learning granularity and the enhancement for network topology change. We compare IFS-RL to state-of-the-art forwarding strategies in ndnSIM. Experimental results show that IFS-RL can achieve higher throughput and lower packet drop rates.
KW - Forwarding strategy
KW - Learning granularity
KW - Named-Data Networking
KW - Network topology
KW - Reinforcement learning
UR - http://www.scopus.com/inward/record.url?scp=85056878340&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85056878340&partnerID=8YFLogxK
U2 - 10.1145/3229543.3229547
DO - 10.1145/3229543.3229547
M3 - Conference contribution
AN - SCOPUS:85056878340
T3 - NetAI 2018 - Proceedings of the 2018 Workshop on Network Meets AI and ML, Part of SIGCOMM 2018
SP - 54
EP - 59
BT - NetAI 2018 - Proceedings of the 2018 Workshop on Network Meets AI and ML, Part of SIGCOMM 2018
PB - Association for Computing Machinery, Inc
Y2 - 24 August 2018
ER -