Presto: Towards efficient online virtual network embedding in virtualized cloud data centers

doi:10.1016/j.comnet.2016.06.036

Computer Networks

Volume 106, 4 September 2016, Pages 196-208

https://doi.org/10.1016/j.comnet.2016.06.036 Get rights and content

Abstract

As an efficient solution to diversify the future Internet for resource sharing in data centers, the network virtualization enables seamless integration of network experiments, services and architectures with different features by allowing multiple heterogeneous virtual networks (VNs) to simultaneously coexist on a shared substrate infrastructure. Embedding multiple virtual networks onto a shared substrate by allocating substrate resources to virtual nodes and virtual links of VN requests under a collection of constrains is known to be an NP-hard problem even for the offline VN embedding. To deal with this issue, this paper formulates the VN embedding problem as a new multiple objective linear programming optimization program, and solves it in a preemptive strategy by decomposing the problem into node mapping and link mapping phases. Furthermore, based on an Artificial Intelligence resource abstraction model, named Blocking Island (BI), we propose an efficient online heuristic VN embedding algorithm called Presto. Presto operates with quite low computation complexity and greatly reduces the search space, which far outperforms other candidates. The goal of Presto is to maximize the economic revenue of infrastructure providers while minimizing the embedding cost. The extensive simulation results further prove the feasibility and good performance of Presto in revenue, VN request acceptance ratio, computation efficiency and resource utilization.

Introduction

As highly multiplexed shared environments, cloud data centers are equipped with a large number of physical servers and virtual machines (VMs) hosted in servers to simultaneously offer multiple tenants with on-demand use of computing resources in a pay-as-you-go manner [1], [2], [3], [4], [5]. How to efficiently share the physical network resources among multiple tenants that have diversified network topologies with different network characteristics is a key concern. With respect to this issue, network virtualization has emerged as an efficient technology for resource sharing, where multiple heterogeneous network architectures are allowed to coexist on a shared substrate [6], [7], [8]. Upon on the virtualized shared data centers, the infrastructure providers then make best effort to utilize the substrate resources to serve the users that request customized services with required resources (such as CPU capacities, network bandwidth, etc.) running over different user self-defined network topologies, which are also known as virtual networks (VNs). Each virtual network consists of a set of virtual nodes interconnected through a set of virtual links with required capacities. The allocation of substrate resources to the virtual networks is called virtual network embedding³ (VNE). Each virtual node is mapped onto a substrate node, while each virtual link is mapped onto a substrate path connecting the corresponding substrate nodes under a series of pre-defined constraints. Fig. 1 illustrates an example of virtual network embedding, where two embedded virtual networks share the same substrate network.

The main objective of solving VNE problem is to make efficient use of substrate resources through dynamic and effective VN mapping algorithms. Although embedding diversified virtual networks of different users onto the underlying physical network can maximize the benefits gained from existing hardware of the infrastructure, VNE has been presented as a very challenging resource allocation problem [9], [10], [11] that has been addressed in many research studies [12], [13], [14], [15], [16], [17]. In fact, the VNE problem is NP-hard [13], [15], [16], [18], even in the offline case. Even when all the virtual nodes are embedded, to embed the virtual links is still NP-hard [15], [19], [20]. Naturally, the online case of VNE problem would be more intractable.

In response to this issue, in this paper we propose an efficient online heuristic VNE algorithm, named Presto, based on an Artificial Intelligence resource abstraction model called Blocking Island. Presto decomposes the VNE problem into two separate phases: virtual node mapping and virtual link mapping. In each phase, with the help of BI model, Presto ranks and embeds the virtual nodes or virtual links in a most advantageous order aiming to maximize the acceptance ratio. All the VN requests are also sorted according to some specific metrics, which targets at maximizing the economic revenues and minimizing the embedding cost. In addition, the proposed sliding window based batch processing approach enables Presto with the ability of lookahead and forward checking when processing the dynamically arriving VN requests. Moreover, with the benefit of BI paradigm, the search space is significantly decreased, and accordingly, the computation efficiency is greatly improved.

The primary contributions of this paper can be summarized as below:

1.
Two heuristic variable ordering algorithms HVNO and HVLO are designed.
2.
A node mapping algorithm HNM and a link mapping algorithm HLM are put forward.
3.
To the best of our knowledge, we are the first to apply BI paradigm to solve the VNE problem.
4.
Extensive simulations are conducted to evaluate the performance of Presto.

The rest of the paper is organized as follows. First we briefly review the related research literature in Section 2. Then Section 3 demonstrates the VN mapping model and problem formulation. In Section 4, the formulated multiple objective linear programming problem is presented. Afterwards, Section 5 introduces the BI paradigm. The Presto framework is designed in Section 6 followed by evaluations in Section 7. Section 8 concludes this paper.

Section snippets

Related work and motivation

A considerable number of research and investigations have been conducted on the virtual network embedding problem in recent years. In order to deal with the computationally intractable VNE problem, most of the proposals resort to heuristic algorithms aiming to find some feasible solutions other than optimal solutions. Generally, the existing works can be classified into two categories: offline algorithms [14], [21], [22] and online algorithms [12], [13], [16], [18], [23].

The offline algorithms

Substrate network model

The underlying substrate network can be modelled as a weighted undirected graph $G_{s} = (N_{s}, E_{s}),$ where N_s is the set of substrate nodes of the network and E_s is the set of substrate links between nodes of N_s. Each substrate node n_s ∈ N_s has an associated capacity weight value C(n_s) to denote the available CPU capacity of the physical node n_s. Each substrate link e_s(i, j) ∈ E_s between two substrate nodes i and j is associated with a bandwidth capacity B(e_s), which denotes the amount of available

Multiple objectives linear programming optimization model

In this section, the virtual network embedding problem is formulated as a Multiple Objectives Linear Programming (MOLP) optimization problem. The goal is to find some feasible solutions which aim to optimize the two given objectives while satisfying a series of constraints including capacity constraints, flow constraints, domain constraints and some binary constraints.

Blocking island paradigm

As a resource abstraction model derived from Artificial Intelligence, Blocking Island (BI) was firstly proposed by Christian Frei, et al. in [35] to represent the availability of network link bandwidth. BI is defined as: A β-Blocking Island (β-BI) for a node x is the set of all nodes of the network that can be reached from x using links with at least β available resources, including x. In addition to link bandwidth abstraction, in this paper we extend the BI model to abstract the available CPU

Presto: BI-based online heuristic virtual network embedding framework

As aforementioned, the virtual network embedding (VNE) is an NP-hard problem and the traditional approaches suffer from a high complexity due to the huge searching space consisting of exponential number of nodes and routes. In response to this issue, with the benefit of BI model, we propose an efficient BI based heuristic framework, named Presto, to solve the online virtual network embedding problem with much lower and more manageable complexity.

Evaluation

In order to evaluate the performance of Presto, we implemented a Presto prototype by extending our DCNSim simulator [38] with some accompanying Python scripts. We firstly give an overview on the simulation environment. Then we demonstrate the evaluation results of Presto with respect to some performance metrics including acceptance ratio, gained revenue, embedding cost and computation efficiency.

Conclusion and future work

This paper aims to achieve an efficient online virtual network embedding algorithm in virtualized cloud data centers. In order to deal with the computationally intractable VNE problem, which is known as NP-hard, we formulated it as an MOLP problem with multiple practical objectives and designed an efficient VNE framework Presto consisting of a series of heuristic algorithms such as DHRO, HVNO, HVLO, HNM and HLM. With the benefit of Blocking Island paradigm derived from an AI model, Presto

References (40)

WangL. et al.
Harnessing frequency domain for cooperative sensing and multi-channel contention in crahns
Wireless Commun. IEEE Trans.
(2014)
N.M.K. Chowdhury et al.
Vineyard: Virtual network embedding algorithms with coordinated node and link mapping
IEEE/ACM Trans. Netw. (TON)
(2012)
LiangZ. et al.
A double-module immune algorithm for multi-objective optimization problems
Appl. Soft Comput.
(2015)
LiJ. et al.
Online optimization for scheduling preemptable tasks on iaas cloud systems
J. Parallel Distrib. Comput.
(2012)
C. Papagianni et al.
On the optimal allocation of virtual resources in cloud computing networks
Comput. IEEE Trans.
(2013)
J. Dean et al.
Mapreduce: simplified data processing on large clusters
Commun. ACM
(2008)
WangT. et al.
Rethinking the data center networking: Architecture, network protocols, and resource sharing
Acc. IEEE
(2014)
YanQ. et al.
Software-defined networking (sdn) and distributed denial of service (ddos) attacks in cloud computing environments: A survey, some research issues, and challenges
IEEE Commun. Surv. Tut.
(2015)
WangT. et al.
Jielin: A scalable and cost-effective server-centric data center network architecture
2015 IEEE Global Communications Conference (GLOBECOM)
(2015)
WangT. et al.
Improving the efficiency of server-centric data center network architectures
Communications (ICC), 2014 IEEE International Conference on
(2014)

WangT. et al.

Novacube: A low latency torus-based network architecture for data centers

Global Communications Conference (GLOBECOM), 2014 IEEE

(2014)

T. Anderson et al.

Overcoming the internet impasse through virtualization

Computer

(2005)

M.F. Bari et al.

Data center network virtualization: A survey

Commun. Surv. Tut. IEEE

(2013)

A. Fischer et al.

Virtual network embedding: A survey

Commun. Surv. Tut. IEEE

(2013)

G. Su et al.

Qos constrained optimization of cell association and resource allocation for load balancing in downlink heterogeneous cellular networks

Ksii Trans. Internet Inf. Syst.

(2015)

LiJ. et al.

Resource allocation robustness in multi-core embedded systems with inaccurate information

J. Syst. Arch.

(2011)

ChengX. et al.

Virtual network embedding through topology awareness and optimization

Comput. Netw.

(2012)

LuJ. et al.

Efficient mapping of virtual networks onto a shared substrate

Washington University in St. Louis, Tech. Rep

(2006)

M.R. Rahman et al.

Survivable virtual network embedding

NETWORKING 2010

(2010)

YuM. et al.

Rethinking virtual network embedding: substrate support for path splitting and migration

ACM SIGCOMM Comput. Commun. Rev.

(2008)

Cited by (38)

Towards learning-based energy-efficient online coordinated virtual network embedding framework
2024, Computer Networks
Network virtualization is a highly effective technology for resource sharing within data centers, enabling the coexistence of multiple heterogeneous virtual networks in a shared substrate network, thus achieving resource multiplexing. The efficient embedding of a virtual network onto a substrate network, known as the virtual network embedding (VNE) problem, has been proven to be NP-hard. In response to this challenge, this paper introduces a novel method, named PPO-VNE, which leverages deep reinforcement learning for virtual network embedding. PPO-VNE employs the Proximal Policy Optimization (PPO) algorithm to generate policies and efficiently coordinate node and link mapping. Furthermore, it adopts a hybrid feature extraction approach that combines handcrafted features with features extracted using graph convolutional networks. The proposed reward function takes multiple objectives into account, guiding the learning process. We implemented a prototype of PPO-VNE and conducted experiments based on the simulation environment, in which the substrate network has 100 nodes, with a probability of 0.1 generating edges between any two node, and eventually there will be about 500 physical links. We evaluate the performance of our PPO-VNE approach from the perspective of overall acceptance rate, overall revenue, revenue-to-cost ratio, maximum energy consumption per unit time and revenue-energy consumption coefficient. Comprehensive simulation results in different scenarios show that our PPO-VNE approach achieves the superior performance on most metrics compared with the existing state-of-the-art approaches, where the overall acceptance rate, overall revenue, revenue-energy consumption coefficient are increased by up to 6.4%, 21.3% and 41.4%, respectively, and the maximum energy consumption per unit time are reduced by up to 22.1%.
EE-CTA: Energy efficient, concurrent and topology-aware virtual network embedding as a multi-objective optimization problem
2019, Computer Standards and Interfaces
Citation Excerpt :
We implemented four existing algorithms to compare their performance with the proposed EE-CTA. The first algorithm is Presto (2016) which uses preemptive strategy with making a Blocking Island tree for the substrate network [17]. The second compared algorithm is ACO (2016) which divides each virtual network to ring or tree topology and then uses two strategies to embed divided subsections [2].
Network virtualization provides the substruction of various heterogeneous Virtual Networks (VNs) on a single physical infrastructure which is called Virtual Network Embedding (VNE) and is known as an Np-hard problem. The VNE includes two sub-problem; virtual node mapping and virtual link mapping. Related works do not consider network topology and energy efficiency in the embedding process. This paper proposes Energy Efficient, Concurrent, and Topology-Aware (EE-CTA) algorithm as a new concurrent VNE method. Also, EE-CTA is energy-efficient due to using servers status and renewable energy resources when they are available. Our proposed EE-CTA has focused on network topology with assigning reachability rank to resources. In order to achieve all of these goals, we model VNE as a multi-objective optimization problem and solve it by Non-dominated Sorting Genetic Algorithm (NSGA-II). We compare EE-CTA with Presto, Ant Colony Optimization (ACO), Topology and Migration-Aware Energy Efficient (TMAE), and RW-Max match methods. The evaluation results demonstrate our method improves revenue, acceptance ratio, cost, and energy usage.
Green virtual network embedding with supervised self-organizing map
2019, Neurocomputing
Citation Excerpt :
According to the studies, virtual network embedding problem has two phases of node mapping and link mapping [18]. In the node mapping phase, each virtual node is mapped on a physical/substrate node with sufficient empty capacity [19]. In the link mapping phase, each virtual link is mapped on one or multiple disjoint paths with sufficient bandwidth.
Virtualization in the data center network is used to overcome the resistance of Internet ossification. Node mapping in virtual network embedding (VNE) process is more important than link mapping because each node mapping has a lot of choices with high probability. Historical data always have valuable information which displays how to be mapped virtual networks with special features. Previous works use physical network historical information, and they do not pay attention to the characteristics of virtual networks during the embedding process. To address this problem, this paper proposes SoGVNE (Self-organized Green Virtual Network Embedding) which uses previous successful virtual node mapping information to train a self-organizing map (SOM) neural network. Then, it will respond to future requests quickly and with fewer resources. SoGVNE uses virtual networks historical data and is able to embed more than one virtual network concurrently. Using supervised SOM makes the SoGVNE aware of resources with renewable energy and uses them to manage energy consumption. This paper evaluates SoGVNE with reinforcement-learning, Monte-Carlo and Presto methods. The results indicate the high efficiency of the proposed method in learning the node mapping process and achieving high profit and acceptance ratio with low cost.
Towards full virtualization of SDN infrastructure
2018, Computer Networks
Citation Excerpt :
Network virtualization draws more and more attention in recent years. In public cloud, tenants outsource their infrastructure to the cloud data center, and each tenant has the specific requirement for his virtual network in the cloud [2]. In the wide-area network, people are thinking on network slicing [28], so as to support multiple logical (virtualized) network infrastructure, such as phone network and IoT network, in the future 5G core network [3,4].
With the emerging multi-tenant virtual networks in the cloud data center and the network slicing in wide-area network, full virtualization of the hardware network infrastructure is required. In this work, we propose SVirt, a full virtualization architecture of the software defined networking (SDN) infrastructure, which can flexibly support multiple virtual networks (tenants) to simultaneously run SDN on top of the same physical network. In order to simultaneously run multiple virtual SDN switches with different processing logics on the same physical switch, we redesign the physical switch’s processing pipeline with a “late-binding key extractor (LBX)”. In the control plane, SVirt enables “many-to-one” and “one-to-many” mapping when allocating the physical resource for a virtual network, which embraces arbitrary topology and ternary content-addressable memory (TCAM) resource demanded by a virtual network. In the data plane, SVirt explicitly carries the forwarding context information in the packets, overcoming the “context-loss problem” in a virtual network. We develop a NetFPGA prototype of SVirt switch. Event-driven simulations and prototype-based experiments demonstrate that, compared with traditional approaches, SVirt significantly enhances the cloud’s capability to accept various virtual SDN requests with resource utilization ratio of more than 90% and improves the network’s throughput by around 236%. Besides, SVirt can hold more flows than traditional approaches, thus reducing two-order more table misses.
Incorporating energy and load balance into virtual network embedding process
2018, Computer Communications
Citation Excerpt :
In the subsequent section, we will introduce these two kinds of virtual network embedding algorithms. In recent years, energy-aware virtual network embedding algorithms [1–5] have received considerable attention because the power consumption of an average data center is the same as 25,000 households [6] and it is becoming the major energy cost of network operators which almostly account for 42% of the total operating costs. In addition, the security vulnerabilities caused by network virtualization technologies have been attracted great attention from researchers.
Virtual network embedding has propounded as one of the most important techniques to address the issue of Internet architecture ossification. Previous studies on virtual network embedding algorithms focus primarily on the revenue of service providers (SPs) and the acceptance ratio of virtual network requests (VNRs), regardless of the energy consumption for substrate nodes and substrate links. To deal with this issue, some of researchers proposed energy-aware virtual network embedding algorithm to do a tradeoff between the revenue and energy consumption. However, the problem of virtual network embedding is a multi-objective optimization problem, which should take into account the multiple factors such as revenue, energy consumption, quality of services (QoS), load balancing of nodes and links, and so on. In this paper, we incorporate node energy consumption and node resource utilization into the embedding potential computation for substrate nodes, and propose a comprehensive node ranking method called CNRM to measure the substrate node importance. In addition, we utilize an improved differentiated pricing strategy to assign different weights to substrate links according to their resource utilization ratios, adopt the shortest path algorithm to select substrate path with the minimum energy consumption for virtual link during the link mapping process. Extensive simulation results demonstrated that our algorithm achieves a better tradeoff among the acceptance ratio, revenue/cost ratio, energy consumption and network load balance.
Towards Learning-Based Energy-Efficient Online Coordinated Virtual Network Embedding Framework
2023, SSRN

View all citing articles on Scopus

Ting Wang received the Ph.D. degree in Computer Science and Engineering from Hong Kong University of Science and Technology, Hong Kong SAR China, in 2015, the Master Eng. degree from Warsaw University of Technology, Poland, in 2011, and the Bachelor Sci. degree from University of Science and Technology Beijing, China, in 2008. From 02.2012 to 08.2012 he visited as a research assistant in the Institute of Computing Technology, Chinese Academy of Sciences, China. He is currently a research scientist in Alcatel Lucent Bell Labs, Shanghai, China. His research interests include data center networks, cloud computing, green computing, network function virtualization, software defined network and next generation networks.

Mounir Hamdi received the B.S. degree in Electrical Engineering – Computer Engineering minor (with distinction) from the University of Louisiana in 1985, and the MS and the PhD degrees in Electrical Engineering from the University of Pitts-burgh in 1987 and 1991, respectively. He is a Chair Professor at the Hong Kong University of Science and Technology, and was the head of department of computer science and engineering. Now he is the Dean of the College of Science, Engineering and Technology at the Hamad Bin Khalifa University, Qatar. He is an IEEE Fellow for contributions to design and analysis of high-speed packet switching. He is/was on the Editorial Board of various prestigious journals and magazines including IEEE Transactions on Communications, IEEE Communication Magazine, Computer Networks, Wireless Communications and Mobile Computing, and Parallel Computing as well as a guest editor of IEEE Communications Magazine, guest editor-in-chief of two special issues of IEEE Journal on Selected Areas of Communications, and a guest editor of Optical Networks Magazine. He has chaired more than 20 international conferences and workshops including The IEEE International High Performance Switching and Routing Conference, the IEEE GLOBECOM/ICC Optical networking workshop, the IEEE ICC High-speed Access Workshop, and the IEEE IPPS HiNets Workshop, and has been on the program committees of more than 200 international conferences and workshops. He was the Chair of IEEE Communications Society Technical Committee on Transmissions, Access and Optical Systems, and Vice-Chair of the Optical Networking Technical Committee, as well as member of the ComSoc technical activities council. He received the best paper award at the IEEE International Conference on Communications in 2009 and the IEEE International Conference on Information and Networking in 1998. He also supervised the best PhD paper award among all universities in Hong Kong.

¹: Student Member, IEEE

²: Fellow, IEEE

View full text

Presto: Towards efficient online virtual network embedding in virtualized cloud data centers

Abstract

Introduction

Section snippets

Related work and motivation

Substrate network model

Multiple objectives linear programming optimization model

Blocking island paradigm

Presto: BI-based online heuristic virtual network embedding framework

Evaluation

Conclusion and future work

Wireless Commun. IEEE Trans.

IEEE/ACM Trans. Netw. (TON)

Appl. Soft Comput.

J. Parallel Distrib. Comput.

Comput. IEEE Trans.

Commun. ACM

Rethinking the data center networking: Architecture, network protocols, and resource sharing

Acc. IEEE

Software-defined networking (sdn) and distributed denial of service (ddos) attacks in cloud computing environments: A survey, some research issues, and challenges

IEEE Commun. Surv. Tut.

Jielin: A scalable and cost-effective server-centric data center network architecture

2015 IEEE Global Communications Conference (GLOBECOM)

Improving the efficiency of server-centric data center network architectures

Communications (ICC), 2014 IEEE International Conference on

Novacube: A low latency torus-based network architecture for data centers

Global Communications Conference (GLOBECOM), 2014 IEEE

Overcoming the internet impasse through virtualization

Computer

Data center network virtualization: A survey

Commun. Surv. Tut. IEEE

Virtual network embedding: A survey

Commun. Surv. Tut. IEEE

Qos constrained optimization of cell association and resource allocation for load balancing in downlink heterogeneous cellular networks

Ksii Trans. Internet Inf. Syst.

Resource allocation robustness in multi-core embedded systems with inaccurate information

J. Syst. Arch.

Virtual network embedding through topology awareness and optimization

Comput. Netw.

Efficient mapping of virtual networks onto a shared substrate

Washington University in St. Louis, Tech. Rep

Survivable virtual network embedding

NETWORKING 2010

Rethinking virtual network embedding: substrate support for path splitting and migration

ACM SIGCOMM Comput. Commun. Rev.