Monte Carlo Tree Search with Last-Good-Reply Policy for Cognitive Optimization of Cloud-Ready Optical Networks

  • PDF / 1,624,551 Bytes
  • 23 Pages / 439.37 x 666.142 pts Page_size
  • 79 Downloads / 166 Views

DOWNLOAD

REPORT


Monte Carlo Tree Search with Last‑Good‑Reply Policy for Cognitive Optimization of Cloud‑Ready Optical Networks Michal Aibin1   · Krzysztof Walkowiak2 Received: 23 February 2020 / Revised: 17 May 2020 / Accepted: 9 July 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020

Abstract The rapid development of Cloud Computing and Content Delivery Networks (CDNs) brings a significant increase in data transfers that leads to new optimization challenges in inter-data center networks. In this article, we focus on the cross-stratum optimization of an inter-data center Elastic Optical Network (EON). We develop an optimization approach that employs machine learning Monte Carlo Tree Search (MCTS) algorithm for the simulation of future traffic to improve the performance of the network regarding the request blocking and the operational cost. The key novelty of our approach is using various selection strategies applied to the phase of building a search tree under different network scenarios. We evaluate the performance of these selection strategies using representative topologies and real-data provided by Amazon Web Services. The main conclusion is that the approach based on the policy of Last-GoodReply with Forgetting enables more efficient cloud resource allocation, which results in lower request blocking, thus, reduces the operational cost of the network. Keywords  Elastic optical networks · Dynamic routing · Cloud services · Traffic prediction · Machine learning

1 Introduction From streaming platforms to self-driving cars cloud computing and content-oriented services are getting more and more popular. With the growth of these services, there is an increased demand for big data transfers [1]. Currently, the cloud * Michal Aibin [email protected] Krzysztof Walkowiak [email protected] 1

Department of Computing, British Columbia Institute of Technology, Vancouver, BC, Canada

2

Department of Systems and Computer Networks, Wroclaw University of Science and Technology, Wroclaw, Poland



13

Vol.:(0123456789)



Journal of Network and Systems Management

data centers (DCs) are no longer a new thing—they become to be a standard resource, used by many companies. Everything is measured by the use of virtual resources and payable per hours of using it. In this paper, we consider Cross-Stratum Optimization (CSO) for the provisioning of DC services in optical networks, as a centralized network control becomes more critical in hybrid and complex network architectures [2, 3]. Not only are cloud solutions under the revolution. The current optical technology, which is Dense Wavelength Division Multiplexing (DWDM), is being replaced shortly by a more promising, more flexible and more resilient solution—Elastic Optical Networks (EONs). In contrast to the 50 GHz fixed grid used in traditional DWDM, EONs using a frequency grid with 12.5 GHz granularity better utilize the spectrum resources [4]. The results from [5] clearly show that the EONs outperforms conventional DWDM networks in the context of network resil