[Paper Review] Machine Learning Empowered Intelligent Data Center Networking: A Survey

Bo Li, Ting Wang+4 more|arXiv (Cornell University)|Feb 28, 2022

Cloud Computing and Resource Management4 citations

TL;DR

This survey proposes a comprehensive analysis of machine learning (ML)-driven intelligent data center networking, covering flow prediction, classification, load balancing, resource management, routing, and congestion control. It introduces the REBEL-3S framework to objectively evaluate ML-based solutions and identifies key challenges and future research directions in achieving self-optimizing, adaptive data center networks.

ABSTRACT

To support the needs of ever-growing cloud-based services, the number of servers and network devices in data centers is increasing exponentially, which in turn results in high complexities and difficulties in network optimization. To address these challenges, both academia and industry turn to artificial intelligence technology to realize network intelligence. To this end, a considerable number of novel and creative machine learning-based (ML-based) research works have been put forward in recent few years. Nevertheless, there are still enormous challenges faced by the intelligent optimization of data center networks (DCNs), especially in the scenario of online real-time dynamic processing of massive heterogeneous services and traffic data. To best of our knowledge, there is a lack of systematic and original comprehensively investigations with in-depth analysis on intelligent DCN. To this end, in this paper, we comprehensively investigate the application of machine learning to data center networking, and provide a general overview and in-depth analysis of the recent works, covering flow prediction, flow classification, load balancing, resource management, routing optimization, and congestion control. In order to provide a multi-dimensional and multi-perspective comparison of various solutions, we design a quality assessment criteria called REBEL-3S to impartially measure the strengths and weaknesses of these research works. Moreover, we also present unique insights into the technology evolution of the fusion of data center network and machine learning, together with some challenges and potential future research opportunities.

Motivation & Objective

• Address the growing complexity and dynamic nature of modern data center networks (DCNs) due to exponential scale-up and heterogeneous traffic demands.
• Systematically review and analyze recent ML-based research in DCN optimization across key areas including flow prediction, classification, load balancing, resource management, routing, and congestion control.
• Propose a novel, multi-dimensional evaluation framework—REBEL-3S—to objectively assess the strengths and weaknesses of existing ML-based DCN solutions.
• Identify open challenges and future research opportunities in integrating ML with DCNs, particularly in real-time, dynamic, and secure network intelligence.

Proposed method

• Conduct a comprehensive survey of recent ML-based works in data center networking, focusing on six core application areas: flow prediction, flow classification, load balancing, resource management, routing optimization, and congestion control.
• Design the REBEL-3S quality assessment framework—comprising Relevance, Efficiency, Bypassing, Evaluation, Learning, and Scalability—to provide a standardized, impartial evaluation of ML-based DCN solutions.
• Analyze the impact of data collection strategies, including traffic overhead, computational cost, and privacy risks, on the effectiveness of ML models in DCNs.
• Examine the limitations of current communication protocols in handling heterogeneous, real-time, and bursty workloads, and highlight the need for ML-enhanced, adaptive protocols.
• Investigate network visualization challenges such as routing invisibility, end-to-end pipeline opacity, and QoS opacity, and propose ML-based solutions to improve fault diagnosis and O&M efficiency.
• Synthesize insights on the evolution of AI integration in DCNs and identify under-explored research directions, including intelligent protocol design and secure, scalable deployment.

Experimental results

Research questions

• RQ1 How can machine learning effectively address the challenges of dynamic, heterogeneous, and high-volume traffic in modern data center networks?
• RQ2 What are the key performance and design trade-offs in existing ML-based solutions for DCN optimization across different application areas?
• RQ3 To what extent do current ML-based DCN solutions achieve real-time adaptability, scalability, and operational efficiency in production environments?
• RQ4 How can a standardized, multi-dimensional evaluation framework be designed to fairly compare diverse ML-based DCN approaches?
• RQ5 What are the major open challenges and promising future research directions in building truly intelligent, self-optimizing data center networks?

Key findings

• The REBEL-3S framework provides a systematic, impartial, and multi-perspective evaluation mechanism to assess ML-based DCN solutions across six critical dimensions: Relevance, Efficiency, Bypassing, Evaluation, Learning, and Scalability.
• Despite significant progress, current ML-based DCN solutions still face challenges in real-time processing, data collection overhead, and security, particularly in dynamic and heterogeneous service environments.
• Network visualization remains a critical gap in current O&M practices, with issues like routing invisibility and end-to-end pipeline opacity severely hindering fault diagnosis and performance monitoring.
• Existing communication protocols fail to meet the demands of modern DCNs in terms of compatibility with legacy systems and adaptability to micro-bursts, indicating a need for ML-augmented protocol design.
• The integration of AI into data center networks is still in its early stages, with limited maturity in practical deployment, especially in areas like intelligent congestion control and adaptive routing.
• Future research should focus on developing secure, scalable, and low-overhead ML models that can enable autonomous, intent-driven, and self-healing data center networks.