Elsevier

Computer Networks

Volume 193, 5 July 2021, 108104
Computer Networks

Application of a Long Short Term Memory neural predictor with asymmetric loss function for the resource allocation in NFV network architectures

https://doi.org/10.1016/j.comnet.2021.108104Get rights and content

Abstract

Traffic and cloud resource prediction methodologies have been recently used in Network Function Virtualization environment for cloud and bandwidth resource allocation purposes. Both traditional and innovative prediction methodologies have been proposed for the application of allocation procedures. For instance Long Short Term Memory-based prediction techniques have been shown to be very effectiveness to allocate the resources. All of these techniques are based on the minimization of a symmetric cost function as the Root Mean Square Error that equally weights positive and negative prediction errors. However the error sign can differently impact the cost increase due to prediction errors. For instance when the Quality of Service degradation cost due to traffic loss is prevalent with respect to the cloud resource allocation cost, an algorithm is preferable that overestimates the offered traffic; conversely the traffic underestimation is preferable in the opposite case when the cloud allocation cost is higher than the QoS degradation one. For this reason we propose an Asymmetric LSTM traffic prediction procedure in which the cost function is defined so as to take into account both the QoS degradation and cloud resource allocation costs. In a typical network and traffic scenario, we show how the proposed solution allows for cost decrease by 40% with respect to classical LSTM prediction methodology based on the Root Mean Square Error.

Introduction

Network Function Virtualization (NFV) utilizes the virtualization technologies to consolidate various types of network appliances onto standard high-volume off-the-shelf servers. These technologies bring significant advantages to service providers including reducing the capital and operation expenditures such as equipment costs and energy bill, and also improving the flexibility and innovation in the offered services. The introduction of NFV [1], [2] technology allows the implementation of software middleboxes located in data centers, referred to as Network Function Virtual Infrastructure-Point of Presence (NFVI-PoP) and running on virtual machines. The NFV architecture defines Service Function Chain (SFC) which is a set of Service Functions (SF) to be performed according to a given order. In these last few years the problem of resource allocation and routing of SFCs has been widely investigated [3], [4], [5], [6], [7], [8], [9], [10]. NFV is a very flexible technology that allows the network operator to manage the traffic variations with a re-allocation of computational and memory resources and taking advantage of the migration of virtual machines [11], [12]. The reconfiguration of cloud and bandwidth resources in various traffic scenarios has been extensively studied and investigated in the literature. The studies focused on reactive techniques based on which the network is reconfigured as soon as traffic changes occur [13], [14], [15], [16], [17], [18], [19]. Everyone agrees that reactive techniques are ineffective in relation to the high variability of traffic. For this reason proactive reconfiguration techniques have been proposed where the traffic or the amount of resources needed is predicted [20]. Most of the techniques proposed in the literature are based on Artificial Intelligence (AI) [21]. For example, techniques based on traffic estimation aim to estimate exactly the daily traffic values in order to activate a reconfiguration of bandwidth and cloud resources minimizing a performance index (power consumption, cost of operation, etc.). It is well known that no traffic estimation technique is able to estimate traffic values exactly. There is always an innovative and non-predictable traffic component. Usually the prediction is based on minimizing the Root Mean Square Error (RMSE) which is a symmetrical cost function. Unfortunately, the RMSE-based prediction is not effective because it weighs the positive and negative traffic estimation errors in the same way. Conversely, the sign of the estimation error may have a different impact on the performance index that is minimized when resources are reconfigured. For example, if the index to be optimized is the network operation cost we could have that: (i) an overestimation of traffic could lead to an over-provisioning cost due to the cost of additional cloud resources used compared to those needed; (ii) an underestimation of traffic could lead to an under-provisioning cost due to the price paid by the network operator for the consequent degradation of Quality of Service (QoS) caused by the undersizing of resources. The costs of over-provisioning and under-provisioning may be of different value and therefore a traffic minimization based on a symmetric error function fails to minimize the network operation cost.

To our best knowledge, only Bega et al. [22] propose a solution for mobile network resource orchestration in which the different values of the over-provisioning and under-provisioning costs are taken into account. DeepCog [22] is proposed, a framework for resource allocation to slicing in a 5G mobile environment. It is based on a deep learning technique in which the cost function attributes a rising cost as the amount of over-allocated resources increases and a constant penalty, that is independent of the lost traffic amount, when a QoS degradation occurs.

In this paper we also propose a solution in which over-provisioning and under-provisioning costs are considered and our work differs from [22] in the following points:

  • our solution is proposed for a NFV network scenario where the data center and network model is well detailed and articulated and the application is mainly based on the NFV implementation of middleboxes;

  • traffic forecasting is based on minimizing an asymmetric cost function in the Long Short Term Memory (LSTM) prediction procedure;

  • the cost function is characterized by a single parameter that can be set for a cost penalty not necessarily constant but related to the amount of traffic lost due to under-provisioning.

The paper is organized as follows. The related work is mentioned in Section 2. We describe the problem statement in Section 3. The traffic forecasting technique based on asymmetric loss function is illustrated in Section 4. The numerical results, reported in Section 5, show the effectiveness of the proposed technique with respect to RMSE-based traditional forecasting techniques in an NFV network environment. The conclusions and future research items are described in Section 6.

Section snippets

Related work and research motivation

The SFC deployment problem and its variations have gained a lot of attention during the past few years [23], [24]. The problem of resource allocation has been widely investigated in literature. Solutions have been proposed for static [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25] and dynamic [14], [15] traffic scenarios. In the case of dynamic traffic scenarios, reactive reconfiguration procedures are not adequate due to

Problem statement

The objective of the paper is to propose and evaluate a solution for the cloud and bandwidth resource allocation in NFV environments in which the traffic offered is not a-priory known but it is predicted according to a prediction technique aiming at minimizing the total operational cost. Two cost components are considered: (i) Cloud Resource Allocation Cost; (ii) QoS Degradation Cost occurring when the traffic is incorrectly predicted, less resources are allocated and the Service Provider (SP)

SFC bandwidth forecasting based on LSTM neural networks with asymmetric cost function

The Long Short Term Memory (LSTM) belongs to the family of artificial Recurrent Neural Networks (RNN). RNNs are fundamentally different from traditional feed-forward Artificial Neural Networks (ANN): differently from the feed-forward ANN, the computations in RNN are derived from both current and past inputs and that gives them the ability to capture the temporal correlations between previous information and the current circumstances. Such characteristic of RNNs is ideal for the prediction of

Numerical results

We will evaluate the effectiveness of the asymmetric cost function-based LSTM forecasting model in predicting the requested SFC bandwidths when both the cloud resource allocation and QoS degradation costs are considered. The LSTM forecasting technique will be applied in a real scenario to evaluate the operation cost of an NFV network and compare it to the one achieved when an RMSE traditional forecasting technique is applied.

We describe the simulation environment in Section 5.1. We will show in

Conclusions

We have developed a traffic forecasting algorithm for the allocation of resources in NFV environments that can differently weigh the over-provisioning and under-provisioning costs. The proposed solution is inherited from the classical LSTM prediction algorithm and it is based on minimizing an asymmetric cost function of the prediction error. The use of the prediction technique proposed in an NFV network scenario with the interconnection of four NFVI-PoPs has led to cost advantages by 40%

CRediT authorship contribution statement

Vincenzo Eramo: Conceptualization, Methodology, Writing. Francesco Giacinto Lavacca: Software. Tiziana Catena: Software, Investigation, Validation. Paul Jaime Perez Salazar: Software.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Vincenzo Eramo received his “Laurea” degree in Electronics Engineering in 1995 and his “Dottorato di Ricerca” (Ph.D. degree) in Information and Communications Engineering in 2001, both from the University of Roma “La Sapienza”. From June 1996 to December 1996 he was a researcher at the Scuola Superiore Reiss Romoli. In 1997, he joined the Fondazione Ugo Bordoni as a researcher in the Telecommunication Network Planning group. From November 2002 to October 2005 was an assistant professor and from

References (48)

  • YiB. et al.

    A comprehensive survey of network function virtualization

    Comput. Netw.

    (2018)
  • BarakabitzeA.A. et al.

    5g network slicing using SDN and NFV: A survey of taxonomy, architectures and future challenges

    Comput. Netw.

    (2020)
  • PeiJ. et al.

    Optimal VNF placement via deep reinforcement learning in SDN/NFV-Enabled networks

    IEEE J. Sel. Areas Commun.

    (2020)
  • ChenJ. et al.

    ClusVNFI: A hierarchical clustering-based approach for solving VNFI dilemma in NFV orchestration

    IEEE Access

    (2019)
  • FarkianiB. et al.

    A fast near-optimal approach for energy-aware SFC deployment

    IEEE Trans. Netw. Serv. Manag.

    (2019)
  • El MensoumI. et al.

    MuSC: A multi-stage service chains embedding approach

    J. Netw. Comput. Appl.

    (2020)
  • YuY. et al.

    Network function virtualization resource allocation based on joint benders decomposition and ADMM

    IEEE Trans. Veh. Technol.

    (2020)
  • EramoV. et al.

    Computing and bandwidth resource allocation in multi-provider NFV environment

    IEEE Commun. Lett.

    (2018)
  • Karimzadeh-FarshbafanM. et al.

    Reliability aware service placement using a viterbi-based algorithm

    IEEE Trans. Netw. Serv. Manage.

    (2020)
  • PhamT.M. et al.

    Modeling and analysis of robust service composition for network functions virtualization

    Comput. Netw.

    (2020)
  • EramoV. et al.

    Proposal and investigation of a reconfiguration cost aware policy for resource allocation in multi-provider NFV infrastructures interconnected by elastic optical networks

    J. Lightwave Technol.

    (2019)
  • EramoV. et al.

    Study of reconfiguration cost and energy aware VNE policies in cycle-stationary traffic scenarios

    IEEE J. Sel. Areas Commun.

    (2016)
  • LiuY. et al.

    An approach for service function chain reconfiguration in network function virtualization architectures

    IEEE Access

    (2019)
  • EramoV. et al.

    An approach for service function chain routing and virtual function network instance migration in network function virtualization architectures

    IEEE-ACM Trans. Netw.

    (2017)
  • EramoV. et al.

    Optimizing the cloud resources, bandwidth and deployment costs in multi-providers network function virtualization environment

    IEEE Access

    (2019)
  • FuX. et al.

    Dynamic service function chain embedding for NFV-enabled IoT: A deep reinforcement learning approach

    IEEE Trans. Wireless Commun.

    (2020)
  • Karimzadeh-FarshbafanM. et al.

    A dynamic reliability-aware service placement for network function virtualization (NFV)

    IEEE J. Sel. Areas Commun.

    (2020)
  • YiB. et al.

    Design and implementation of network-aware VNF migration mechanism

    IEEE Access

    (2020)
  • MaW. et al.

    Placing traffic-changing and partially-ordered NFV middleboxes via SDN

    IEEE Trans. Netw. Serv. Manag.

    (2019)
  • H.N. Kim, D. Lee, S. Jeong, H. Choix, J. Yoo, J. Won-Ki Hong, Machine learning-based method for prediction of virtual...
  • S. Schneider, N. Puthenpurayil Satheeschandrany, M. Peuster, H. Karl, Machine learning-based method for prediction of...
  • BegaD. et al.

    DeepCog: Optimizing resource provisioning in network slicing with AI-based capacity forecasting

    IEEE J. Sel. Areas Commun.

    (2020)
  • HantoutiH. et al.

    Traffic steering for service function chaining

    IEEE Commun. Surv. Tutor.

    (2019)
  • LaghrissiA. et al.

    A survey on the placement of virtual resources and virtual network functions

    IEEE Commun. Surv. Tutor.

    (2019)
  • Cited by (0)

    Vincenzo Eramo received his “Laurea” degree in Electronics Engineering in 1995 and his “Dottorato di Ricerca” (Ph.D. degree) in Information and Communications Engineering in 2001, both from the University of Roma “La Sapienza”. From June 1996 to December 1996 he was a researcher at the Scuola Superiore Reiss Romoli. In 1997, he joined the Fondazione Ugo Bordoni as a researcher in the Telecommunication Network Planning group. From November 2002 to October 2005 was an assistant professor and from November 2006 to June 2010 was an Aggregate Professor in the INFOCOM Department of the University of Rome “La Sapienza”. Currently he is an Associate Professor in Department of Engineering of Information, Electronics and Telecommunications. His research activities have been carried out in the framework of national and international projects. In particular Vincenzo Eramo was scientific coordinator for University of Roma “La Sapienza” of E-PhotoONe+ and BONE, two Networks of Excellence focusing on the study of Optical Networks and financed by European Commissions (FP6 and FP7) in 2006–2007 and 2008–2011 respectively. He was an Associate Editor of IEEE Transactions on Computer from July 2011 to June 2015 and he has been an Associate Editor of IEEE Communication Letters since September 2014. He was winner of the Exemplary Editor Award 2016 and 2017 of IEEE Communications Letter.

    Francesco Giacinto Lavacca received the Laurea (M.Sc.) (cum laude) degree in electronic engineering in 2013 and the Ph.D. degree in Information Technology in 2017 from the Sapienza University of Rome, Italy where he is currently a PostDoc researcher at the Department of Information, Electronic and Telecommunication engineering (DIET). In 2016, he has been a Visiting Researcher with the College of Computing, Georgia Institute of Technology, Atlanta, GA, USA. Furthermore, he was involved in the framework of national and international projects, like AAA (Advanced Avionic Architecture) and NMLV (Nano Micro Launch Vehicle) with ASI (Italian Space Agency) and ESA (European Space Agency) respectively. His current research interests are in the fields of all-optical networks and switching architectures, 5G networks, Network Function Virtualization, Time-Triggered and deterministic Ethernet.

    Tiziana Catena received the Laurea (M.Sc.) (cum laude) degree in electronic engineering in 2018 from Sapienza University of Rome, Italy. She is following a doctorate course in INFORMATION AND COMMUNICATION TECHNOLOGY (ICT) at Sapienza University of Rome. His current research interests are in the fields of Network Function Virtualization.

    Paul Jaime Perez Salazar received the Laurea (M.Sc.) (cum laude) degree in aerospace engineering in 2020 from Sapienza University of Rome, Italy. Currently he is involved in research activities on Network Function Virtualization in University of Roma Sapienza.

    View full text