Periodic Reporting for period 1 - OTIS (Optimal Orchestration of Virtualized Network Functions for Improved Quality-of-Service)
Période du rapport: 2022-07-01 au 2024-06-30
The problem being addressed: This project focuses on Service Function Chain (SFC) parallelism design and efficient VNF placement to boost the NFV performance and improve QoS for telecom networks.
The importance for society: Telecom networks provide vital communication infrastructure and play a key role in promoting the economic and social well-being. Network Function Virtualization (NFV) has been identified as a promising and cost-efficient solution to reduce capital expenditure (CAPEX) and operational expenses (OPEX), by decoupling traditional NFs from the proprietary equipment and running them as software-based virtualized instances, known as Virtual Network Functions (VNFs). The technologies developed in this project can serve as the theoretic tools and potential technical solutions to enhance NFV performance, and minimize OPEX and CAPEX for future network applications to address critical issues related to European and global industries such as autonomous driving, online treatment, and smart city, and hence bring benefits for the society.
The overall objectives: This project aims to develop a suite of optimal VNF orchestration schemes to achieve higher NFV performance and improve the QoS of telecom networks. The specific objectives for the project “OTIS” are listed as follows: (1) establish network function chain parallelism paradigms; (2) design optimal virtualized network function placement approaches; (3) systems evaluation, prototype system and demonstration for the optimal orchestration of VNFs in NFV-based telecom networks.
The importance for society: Telecom networks provide vital communication infrastructure and play a key role in promoting the economic and social well-being. Network Function Virtualization (NFV) has been identified as a promising and cost-efficient solution to reduce capital expenditure (CAPEX) and operational expenses (OPEX), by decoupling traditional NFs from the proprietary equipment and running them as software-based virtualized instances, known as Virtual Network Functions (VNFs). The technologies developed in this project can serve as the theoretic tools and potential technical solutions to enhance NFV performance, and minimize OPEX and CAPEX for future network applications to address critical issues related to European and global industries such as autonomous driving, online treatment, and smart city, and hence bring benefits for the society.
The overall objectives: This project aims to develop a suite of optimal VNF orchestration schemes to achieve higher NFV performance and improve the QoS of telecom networks. The specific objectives for the project “OTIS” are listed as follows: (1) establish network function chain parallelism paradigms; (2) design optimal virtualized network function placement approaches; (3) systems evaluation, prototype system and demonstration for the optimal orchestration of VNFs in NFV-based telecom networks.
The project has fully achieved its objectives and milestones for the period. In details,
Task 1 focused on FDRL-based partial VNF parallelism that explores the optimal parallel placement of VNFs via FL and DRL to reduce the end-to-end latency of SFCs. Specifically, we first designed three parallelism rules to enumerate all the feasible configurations of partial VNF parallelism based on the dependency of each VNF pair. Then, we allocated the whole SFCs into all domains, considering the available resources of each domain and potential parallel VNF pairs that should be assigned to the same domain to reduce inter-domain information ex-change. Finally, a parallel VNF placement model that identifies the minimum end-to-end latency with the designed parallelism principles was trained by domain orchestrators and the cloud server, through latency/reward-based federated aggregation.
Related research results have been published in IEEE/ACM Transactions on Networking.
H. Huang, J. Tian, G. Min, H. Yin, C. Zeng, Y. Zhao, and D. Wu, “Parallel Placement of Virtualized Network Functions via Federated Deep Reinforcement Learning”, IEEE/ACM Transactions on Networking, 2024, 32(2):2936-2949.
Task 2 was devoted to effective resource and QoS aware SFC placement. The SFC placement was modeled as a Markov-chain-based optimization problem with the Markov property of its VNFs, where the set of all possible placement states on diverse nodes was regarded as a state space in the Markov chain and each state was jointly determined by the initial state and transition matrix. On this basis, the SFCs associated with traffic requests were re-sorted so as to efficiently instantiate VNFs of the same type. Besides, a Backward-Viterbi-based heuristic algorithm was presented to conduct the optimal VNF placement in Markov chain space to reduce resource consumption, followed by the QoS-based instantiation of virtual links between adjacent VNFs.
Related research results have been published in IEEE Transactions on Services Computing.
H. Huang, J. Tian, H. Yin, G. Min, D. Wu, and W. Miao, “RQAP: Resource and QoS Aware Placement of Service Function Chains in NFV-Enabled Networks”, IEEE Transactions on Services Computing, 2023, 16(6):4526-4539.
Task 3 developed reliability-aware VNF service provisioning, built on multiple-objective Multi-Agent DR (MADRL), to execute VNF orchestration in networks in distributed and parallel manners. Each VNF was hosted in appropriate hard-ware with reliability guarantees, by training reliability-aware models of VNF placement in cross-region networks via MADRL with two alternate complementary goals in two perspectives, i.e. maximizing the reliability and minimizing the failure probability of VNFs, and parallelly instancing VNFs with collaborative multi-agents. To further accelerate the convergence of model training and enhance learning accuracy, the newly-designed training sampling and action exploration were developed.
Related research results have been published in IEEE IEEE/ACM Transactions on Networking.
H. Huang, Y. Cai, G. Min, H. Wang, G. Liu, and D. Wu, “Accurate Prediction of Network Distance via Federated Deep Reinforcement Learning”, IEEE/ACM Transactions on Networking, 2024, 32(2):3301-3314.
H. Huang, J. Tian, Z. Li, G. Min, and H. Wang, “Reliability-Aware Placement of Virtual Network Functions via Multi-Agent Deep Reinforcement Learning”, IEEE/ACM Transactions on Networking (Rejected and resubmitted).
Task 4 mainly adopted the results obtained in Tasks 1, 2 and 3 to develop a software package with two components and an NFV-based experimental testbed prototype. The developed software was devoted to efficient SFC orchestration in NFV-based networks with QoS guarantee. The first component of the software is exploited to execute intelligent SFCP, while the second component is used to resolve the VNFI problem. The built prototype can be adapted to other working scenarios in the broad field of 5G specifications, IoT, and future Internet, etc.
Related research results have been published in IEEE Communications Magazine.
W. Liu, H. Huang, H. Yin, G. Min, Y. Yuan, and D. Wu, “Scalable Blockchain-Based Data Storage in Internet of Things”, IEEE Communications Magazine, 2024, 62(1): 40-45.
Task 1 focused on FDRL-based partial VNF parallelism that explores the optimal parallel placement of VNFs via FL and DRL to reduce the end-to-end latency of SFCs. Specifically, we first designed three parallelism rules to enumerate all the feasible configurations of partial VNF parallelism based on the dependency of each VNF pair. Then, we allocated the whole SFCs into all domains, considering the available resources of each domain and potential parallel VNF pairs that should be assigned to the same domain to reduce inter-domain information ex-change. Finally, a parallel VNF placement model that identifies the minimum end-to-end latency with the designed parallelism principles was trained by domain orchestrators and the cloud server, through latency/reward-based federated aggregation.
Related research results have been published in IEEE/ACM Transactions on Networking.
H. Huang, J. Tian, G. Min, H. Yin, C. Zeng, Y. Zhao, and D. Wu, “Parallel Placement of Virtualized Network Functions via Federated Deep Reinforcement Learning”, IEEE/ACM Transactions on Networking, 2024, 32(2):2936-2949.
Task 2 was devoted to effective resource and QoS aware SFC placement. The SFC placement was modeled as a Markov-chain-based optimization problem with the Markov property of its VNFs, where the set of all possible placement states on diverse nodes was regarded as a state space in the Markov chain and each state was jointly determined by the initial state and transition matrix. On this basis, the SFCs associated with traffic requests were re-sorted so as to efficiently instantiate VNFs of the same type. Besides, a Backward-Viterbi-based heuristic algorithm was presented to conduct the optimal VNF placement in Markov chain space to reduce resource consumption, followed by the QoS-based instantiation of virtual links between adjacent VNFs.
Related research results have been published in IEEE Transactions on Services Computing.
H. Huang, J. Tian, H. Yin, G. Min, D. Wu, and W. Miao, “RQAP: Resource and QoS Aware Placement of Service Function Chains in NFV-Enabled Networks”, IEEE Transactions on Services Computing, 2023, 16(6):4526-4539.
Task 3 developed reliability-aware VNF service provisioning, built on multiple-objective Multi-Agent DR (MADRL), to execute VNF orchestration in networks in distributed and parallel manners. Each VNF was hosted in appropriate hard-ware with reliability guarantees, by training reliability-aware models of VNF placement in cross-region networks via MADRL with two alternate complementary goals in two perspectives, i.e. maximizing the reliability and minimizing the failure probability of VNFs, and parallelly instancing VNFs with collaborative multi-agents. To further accelerate the convergence of model training and enhance learning accuracy, the newly-designed training sampling and action exploration were developed.
Related research results have been published in IEEE IEEE/ACM Transactions on Networking.
H. Huang, Y. Cai, G. Min, H. Wang, G. Liu, and D. Wu, “Accurate Prediction of Network Distance via Federated Deep Reinforcement Learning”, IEEE/ACM Transactions on Networking, 2024, 32(2):3301-3314.
H. Huang, J. Tian, Z. Li, G. Min, and H. Wang, “Reliability-Aware Placement of Virtual Network Functions via Multi-Agent Deep Reinforcement Learning”, IEEE/ACM Transactions on Networking (Rejected and resubmitted).
Task 4 mainly adopted the results obtained in Tasks 1, 2 and 3 to develop a software package with two components and an NFV-based experimental testbed prototype. The developed software was devoted to efficient SFC orchestration in NFV-based networks with QoS guarantee. The first component of the software is exploited to execute intelligent SFCP, while the second component is used to resolve the VNFI problem. The built prototype can be adapted to other working scenarios in the broad field of 5G specifications, IoT, and future Internet, etc.
Related research results have been published in IEEE Communications Magazine.
W. Liu, H. Huang, H. Yin, G. Min, Y. Yuan, and D. Wu, “Scalable Blockchain-Based Data Storage in Internet of Things”, IEEE Communications Magazine, 2024, 62(1): 40-45.
The progress beyond the state-of-the-art:
(1) Three parallelism rules of VNFs have been designed, and furthermore, the inde-pendent NFs have been identified to be instantiated in parallel in networks with ex-plicit consideration of their instance costs.
(2) A FDRL-based partial VNF parallelism is proposed to reduce the end-to-end latency of SFCs.
(3) An effective resource and QoS aware SFC placement is presented to reduce resource consumption and ensure the QoS guarantee of NFV.
(4) The MADRL-based reliability-aware VNF service provisioning is proposed to execute VNF orchestration in NFV-based networks in distributed, parallel and col-laborative manners.
The potential impacts:
The research outcomes achieved in this project developed novel optimal VNF orchestration schemes, by introducing DRL, FDRL and Backward Viterbi heuristic algorithm, to boost the NFV performance and improve QoS for telecom networks, which set a milestone for future European and global studies and raise our visibility in the European academic and industrial circles. In addition to its importance in the wider scientific community, this research will also be of economic significance to related European and global industries such as autonomous driving, online treatment, and smart city.
(1) Three parallelism rules of VNFs have been designed, and furthermore, the inde-pendent NFs have been identified to be instantiated in parallel in networks with ex-plicit consideration of their instance costs.
(2) A FDRL-based partial VNF parallelism is proposed to reduce the end-to-end latency of SFCs.
(3) An effective resource and QoS aware SFC placement is presented to reduce resource consumption and ensure the QoS guarantee of NFV.
(4) The MADRL-based reliability-aware VNF service provisioning is proposed to execute VNF orchestration in NFV-based networks in distributed, parallel and col-laborative manners.
The potential impacts:
The research outcomes achieved in this project developed novel optimal VNF orchestration schemes, by introducing DRL, FDRL and Backward Viterbi heuristic algorithm, to boost the NFV performance and improve QoS for telecom networks, which set a milestone for future European and global studies and raise our visibility in the European academic and industrial circles. In addition to its importance in the wider scientific community, this research will also be of economic significance to related European and global industries such as autonomous driving, online treatment, and smart city.