Path-Link Graph Neural Network for IP Network Performance Prediction

Yangzhe Kong; Dmitry Petrov; Vilho Raisanen; Alexander Ilin

Path-Link Graph Neural Network for IP Network Performance Prediction

Yangzhe Kong, Dmitry Petrov, Vilho Raisanen, Alexander Ilin

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

9 Citations (Scopus)

154 Downloads (Pure)

Abstract

Dynamic resource provisioning and quality assurance for the plethora of end-to-end slices running over 5G and B5G networks require advanced modeling capabilities. Graph Neural Networks (GNN) have already proven their efficiency for network performance prediction. GNN architecture matches well the structures usually met in communications networks. In this paper, the focus is on the IP transport network as one of the end-to-end 5G architecture domains. The recently published RouteNet GNN is taken as a reference and starting point for our study. RouteNet performance is verified by a new implementation in the PyTorch ML library. Next, an alternative Path-Link neural network (PLNet) architecture is proposed and evaluated. After hyper-parameter tuning for both models, the results show that PLNet and RouteNet achieve a similar accuracy level. The advantage of PLNet is in parallel architecture. It is demonstrated that its inference speed is not sensitive to the length of the network's paths.

Original language	English
Title of host publication	2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021)
Editors	Toufik Ahmed, Olivier Festor, Yacine Ghamri-Doudane, Joon-Myung Kang, Alberto E. Schaeffer-Filho, Abdelkader Lahmadi, Edmundo Madeira
Publisher	International Federation for Information Processing (IFIP)
Pages	170-177
Number of pages	8
ISBN (Electronic)	978-3-903176-32-4
Publication status	Published - 17 May 2021
MoE publication type	A4 Conference publication
Event	IEEE International Symposium on Integrated Network Management - Virtual, Online, Bordeaux, France Duration: 17 May 2021 → 21 May 2021 Conference number: 17

Publication series

Name	Integrated network management
ISSN (Print)	1573-0077

Conference

Conference	IEEE International Symposium on Integrated Network Management
Abbreviated title	IM
Country/Territory	France
City	Bordeaux
Period	17/05/2021 → 21/05/2021

Access to Document

Path-Link Graph Neural Network for IP Network Performance PredictionAccepted author manuscript, 2.79 MB

https://ieeexplore.ieee.org/document/9464050

Cite this

Kong, Y., Petrov, D., Raisanen, V., & Ilin, A. (2021). Path-Link Graph Neural Network for IP Network Performance Prediction. In T. Ahmed, O. Festor, Y. Ghamri-Doudane, J.-M. Kang, A. E. Schaeffer-Filho, A. Lahmadi, & E. Madeira (Eds.), 2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021) (pp. 170-177). Article 9464050 (Integrated network management). International Federation for Information Processing (IFIP). https://ieeexplore.ieee.org/document/9464050

Kong, Yangzhe ; Petrov, Dmitry ; Raisanen, Vilho et al. / Path-Link Graph Neural Network for IP Network Performance Prediction. 2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021). editor / Toufik Ahmed ; Olivier Festor ; Yacine Ghamri-Doudane ; Joon-Myung Kang ; Alberto E. Schaeffer-Filho ; Abdelkader Lahmadi ; Edmundo Madeira. International Federation for Information Processing (IFIP), 2021. pp. 170-177 (Integrated network management).

@inproceedings{36c9263c3c8049eba6c6cf1d201555ac,

title = "Path-Link Graph Neural Network for IP Network Performance Prediction",

abstract = "Dynamic resource provisioning and quality assurance for the plethora of end-to-end slices running over 5G and B5G networks require advanced modeling capabilities. Graph Neural Networks (GNN) have already proven their efficiency for network performance prediction. GNN architecture matches well the structures usually met in communications networks. In this paper, the focus is on the IP transport network as one of the end-to-end 5G architecture domains. The recently published RouteNet GNN is taken as a reference and starting point for our study. RouteNet performance is verified by a new implementation in the PyTorch ML library. Next, an alternative Path-Link neural network (PLNet) architecture is proposed and evaluated. After hyper-parameter tuning for both models, the results show that PLNet and RouteNet achieve a similar accuracy level. The advantage of PLNet is in parallel architecture. It is demonstrated that its inference speed is not sensitive to the length of the network's paths.",

author = "Yangzhe Kong and Dmitry Petrov and Vilho Raisanen and Alexander Ilin",

note = "Publisher Copyright: {\textcopyright} 2021 IFIP.; IEEE International Symposium on Integrated Network Management, IM ; Conference date: 17-05-2021 Through 21-05-2021",

year = "2021",

month = may,

day = "17",

language = "English",

series = "Integrated network management",

publisher = "International Federation for Information Processing (IFIP)",

pages = "170--177",

editor = "Toufik Ahmed and Olivier Festor and Yacine Ghamri-Doudane and Joon-Myung Kang and Schaeffer-Filho, {Alberto E.} and Abdelkader Lahmadi and Edmundo Madeira",

booktitle = "2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021)",

address = "Luxembourg",

}

Kong, Y, Petrov, D, Raisanen, V & Ilin, A 2021, Path-Link Graph Neural Network for IP Network Performance Prediction. in T Ahmed, O Festor, Y Ghamri-Doudane, J-M Kang, AE Schaeffer-Filho, A Lahmadi & E Madeira (eds), 2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021)., 9464050, Integrated network management, International Federation for Information Processing (IFIP), pp. 170-177, IEEE International Symposium on Integrated Network Management, Bordeaux, France, 17/05/2021. <https://ieeexplore.ieee.org/document/9464050>

Path-Link Graph Neural Network for IP Network Performance Prediction. / Kong, Yangzhe; Petrov, Dmitry; Raisanen, Vilho et al.
2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021). ed. / Toufik Ahmed; Olivier Festor; Yacine Ghamri-Doudane; Joon-Myung Kang; Alberto E. Schaeffer-Filho; Abdelkader Lahmadi; Edmundo Madeira. International Federation for Information Processing (IFIP), 2021. p. 170-177 9464050 (Integrated network management).

Research output: Chapter in Book/Report/Conference proceeding › Conference article in proceedings › Scientific › peer-review

TY - GEN

T1 - Path-Link Graph Neural Network for IP Network Performance Prediction

AU - Kong, Yangzhe

AU - Petrov, Dmitry

AU - Raisanen, Vilho

AU - Ilin, Alexander

N1 - Conference code: 17

PY - 2021/5/17

Y1 - 2021/5/17

N2 - Dynamic resource provisioning and quality assurance for the plethora of end-to-end slices running over 5G and B5G networks require advanced modeling capabilities. Graph Neural Networks (GNN) have already proven their efficiency for network performance prediction. GNN architecture matches well the structures usually met in communications networks. In this paper, the focus is on the IP transport network as one of the end-to-end 5G architecture domains. The recently published RouteNet GNN is taken as a reference and starting point for our study. RouteNet performance is verified by a new implementation in the PyTorch ML library. Next, an alternative Path-Link neural network (PLNet) architecture is proposed and evaluated. After hyper-parameter tuning for both models, the results show that PLNet and RouteNet achieve a similar accuracy level. The advantage of PLNet is in parallel architecture. It is demonstrated that its inference speed is not sensitive to the length of the network's paths.

AB - Dynamic resource provisioning and quality assurance for the plethora of end-to-end slices running over 5G and B5G networks require advanced modeling capabilities. Graph Neural Networks (GNN) have already proven their efficiency for network performance prediction. GNN architecture matches well the structures usually met in communications networks. In this paper, the focus is on the IP transport network as one of the end-to-end 5G architecture domains. The recently published RouteNet GNN is taken as a reference and starting point for our study. RouteNet performance is verified by a new implementation in the PyTorch ML library. Next, an alternative Path-Link neural network (PLNet) architecture is proposed and evaluated. After hyper-parameter tuning for both models, the results show that PLNet and RouteNet achieve a similar accuracy level. The advantage of PLNet is in parallel architecture. It is demonstrated that its inference speed is not sensitive to the length of the network's paths.

UR - http://www.scopus.com/inward/record.url?scp=85113602999&partnerID=8YFLogxK

M3 - Conference article in proceedings

AN - SCOPUS:85113602999

T3 - Integrated network management

SP - 170

EP - 177

BT - 2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021)

A2 - Ahmed, Toufik

A2 - Festor, Olivier

A2 - Ghamri-Doudane, Yacine

A2 - Kang, Joon-Myung

A2 - Schaeffer-Filho, Alberto E.

A2 - Lahmadi, Abdelkader

A2 - Madeira, Edmundo

PB - International Federation for Information Processing (IFIP)

T2 - IEEE International Symposium on Integrated Network Management

Y2 - 17 May 2021 through 21 May 2021

ER -

Kong Y, Petrov D, Raisanen V, Ilin A. Path-Link Graph Neural Network for IP Network Performance Prediction. In Ahmed T, Festor O, Ghamri-Doudane Y, Kang JM, Schaeffer-Filho AE, Lahmadi A, Madeira E, editors, 2021 IFIP/IEEE INTERNATIONAL SYMPOSIUM ON INTEGRATED NETWORK MANAGEMENT (IM 2021). International Federation for Information Processing (IFIP). 2021. p. 170-177. 9464050. (Integrated network management).

Path-Link Graph Neural Network for IP Network Performance Prediction

Abstract

Publication series

Conference

Access to Document

Other files and links

Fingerprint

Cite this