Back to Search
Start Over
Deep reinforcement learning based multi-layered traffic scheduling scheme in data center networks.
- Source :
-
Wireless Networks (10220038) . Jul2024, Vol. 30 Issue 5, p4133-4144. 12p. - Publication Year :
- 2024
-
Abstract
- A web search, an online video, a connected Nest device, and hundreds of cloud services all give us a response in a fraction of a second. But what really happens when we click search or send a request. The request travels over the public internet and into fiber network. Millions of requests or packets of data travel through miles of cable over land and under sea, converging at one of the many data centers that operate all over the world. Data Center (DC) is the core site of data operation, storage and forwarding, which is also an important part of cloud platform. A large number of commercial switches and servers are usually deployed at the DC. The DC is a complex set of facilities. Data center networks (DCN) is a network applied in the DC, because the traffic in DC presents the typical characteristics of centralized exchange data and increased traffic, which puts forward further requirements for the DCN. DCN connects a large-scale server cluster and is a bridge for data transmission and storage. With the expansion of DC and the increasing number of service types, communication within data centers becomes more frequent. On the other hand, traffic between data centers has also increased dramatically. Considering the multi-layered transmission mode and traffic characteristics of DCN, this paper proposes a software defined networking (SDN) -based multi-layered traffic scheduling scheme for DCN, which is mainly focused on hop count, criticality and cost. Moreover, based on SDN architecture and Deep Q-Network of Reinforcement Learning (RL), the intelligent multi-layered traffic scheduling scheme is proposed to obtain the current optimal global routing strategy according to the real-time traffic demand in the network. The simulation results show that the proposed scheme outperforms benchmarks in terms of average throughput, normalized total throughput, link bandwidth utilization, average round-trip time and network traffic bandwidth loss rate. [ABSTRACT FROM AUTHOR]
Details
- Language :
- English
- ISSN :
- 10220038
- Volume :
- 30
- Issue :
- 5
- Database :
- Academic Search Index
- Journal :
- Wireless Networks (10220038)
- Publication Type :
- Academic Journal
- Accession number :
- 178231170
- Full Text :
- https://doi.org/10.1007/s11276-021-02883-w