Resource management framework for virtual data center embedding based on software defined networking☆
Introduction
With the improvement of informatization level, commerce and other aspects, continuous collision and integration have been happened between all walks of life and the Internet industry. Due to the development of next generation computing model, cloud computing will become the information center of “Internet +” society, allow users to pay on demand and use remote computing, network and storage resources anywhere and anytime. In cloud computing, infrastructure providers (InPs) abstract the infrastructure network resource into a pool. Service providers (SP) can pay for the resources from the InPs and deploy their own service or computing tasks. InPs can implement uniform resource management to improve the utilization rate of physical resources and benefits; SPs just need to focus on business logic without spending a lot of infrastructure cost, maintenance knowledge and cost. As a kind of high efficient computing model, cloud computing has been attracting more and more attention of enterprises and research institutions.
However, the current InPs, such as the Amazon's EC2, mainly provide resources for users in the form of a virtual machine (VM) without providing network isolation and bandwidth guarantee. Networks with TCP/IP protocol stack only provide their best service mode, and the traditional service mode exists some problems in network isolation, security, network performance guarantee, automated management and other aspects [1]. In order to solve these problems, researchers suggest InPs to allocate resources to lessees through a virtual data center (VDC) [1], [2]. The VDC is a virtual resource collection based on the physical network slice, including virtual machines, virtual switches, virtual routers and the virtual links connecting them [2]. Compared with the traditional mode which only provides VM, allocating resources in the way of VDC can achieve the network isolation and bandwidth guarantee.
The implementation of VDC depends on the network virtualization technology. Compared with the fast developed and mature host virtualization technology, the development of network virtualization is relatively backward. The emergence of the concept and the development of software defined networking (SDN) accelerate the process of network virtualization. SDN separates the control plane from the data forwarding plane, and a controller can be used to cooperatively forward components behavior, get traffic statistics data and monitor topology changes [3]. SDN has the characteristics of centralized control, whole network information acquisition and network function virtualization, which can be used to effectively solve the various problems in the data center [4]. At present, the deployment of SDN application gets great development in the data center. Among them, performance and energy saving are two mainly considered aspects [4], [5]. The network virtualization technology based on SDN can effectively and easily establish the VDC on the physical network of a data center, so as to help InPs lease resources in the form of VDC and provide a network isolation and bandwidth guarantee among multiple lessees. As the multiple VDC are logically separate, so SPs can control their own leasehold VDC completely, such as the introduction of custom network protocols and traffic policies.
While providing resources in the form of VDC can solve performance, security and other problems in a data center; but it brings a new challenge of VDC embedding problem at the same time. The VDC embedding problem refers to how to flexibly and effectively request for proper allocation of physical resources for VDC in a data center. An excellent VDC embedding algorithm can help InPs achieve multiple optimization goals, such as the maximization of the utilization rate of resources and benefits, and the minimization of maintenance costs. Although the embedding problem of VDC is similar to the widely studied virtual network (VN), but there are still some differences:
- (1)
The VN embedding problem is mainly oriented to the wide area network (WAN), while the VDC embedding problem is mainly oriented to the resources allocation in a data center;
- (2)
The nodes in VN are the forwarding devices in the WAN, and the VDC contains a variety of nodes, such as hosts, routers, switches, storage nodes and so on [1];
- (3)
For the same request, one physical node in the VN embedding problem can only embed one virtual node [6], while in the cloud computing environment, several VMs of the same VDC can be allocated to one physical host.
This paper mainly discussed the VDC embedding problem in software defined data center. The main contributions are as follows: (1) We propose a framework of data center resources management based on SDN-enabled; (2) We model the VDC embedding problem and analyze the factors that affect InPs’ revenue in the process of VDC embedding; (3) We consider various resource requirements (CPU, memory, bandwidth) and VDC reliability requirements, and propose a new resource allocation algorithm; (4) We conduct the extensive simulations, which shows that the proposed algorithm can achieve higher revenue at a relatively low cost. Finally, based on a dynamic monitoring strategy, the VDC request with a high Revenue/Cost ratio is selected for further optimization algorithm.
The remainder of this paper is organized as follows. We describe problems in Section 2. Section 3 and Section 4 respectively present the problem modeling and the proposed VDC embedding algorithm. Experiments and performance evaluation are introduced in Section 5. Finally, we summarize our work and conclude this paper in Section 6.
Section snippets
VDC embedding problem and framework of resources management in the data center
InPs possess physical infrastructures and earn profits by leasing resources to SPs. SPs need certain infrastructures to deploy their services or tasks. In order to save equipment cost and maintenance cost, SPs can lease resources from InPs in the form of VDC. First of all, when a SP needs to lease resources, it needs to submit a VDC request specification to the InP, including the topology of VDC, the vCPU amount, memory footprint of each virtual node and the minimum bandwidth of each virtual
Network modeling
In this paper, the underlying data center network is expressed as the undirected graph G(N∪X, E), where N denotes the hosts set in the data center, X denotes the switches set in the data center and E denotes the link set between physical nodes(hosts and switches). Each host contains a series of resource properties (such as CPU, memory, etc.), which are expressed as in this paper, where ai represents the ith property of a host. ci(n, t) represents the surplus capacity of
The VDC embedding algorithm based on the topological potential
In order to satisfy the VDC requirements, allocate resources for VDCs at the minimum cost and improve the VDC acceptance ratio and revenue of the algorithm, this paper proposed a heuristic algorithm to solve the VDC embedding problem. The theoretical basis of the algorithm is to use the topological potential to evaluate the importance of nodes in the network, perform the classification of virtual machine nodes in consideration of the VDC reliability demand and try the best to embed the VM with
Simulation environment
We develop a discrete event simulator based on Python to simulate the VDC embedding process in a data center. The data center network uses a FatTree topology with 6 ports, which includes 45 switches and 54 physical hosts. Each host contains 16 vCPUs and a memory of 8096 MB. The bandwidth capacity of each port of the switch is 1000 Mbps. VDC requests are generated by the BRITE [15] topology generator. The VM number of VDC requests evenly distributes between 10 and 50. The vCPU of each VM request
Conclusions
In this paper, we front an important challenge that how to meet the diverse needs of service providers at the minimum cost and maximum profit in the resources allocation by using virtual data centers in which the resources allocation has many advantages. Therefore, we present a novel resource management framework in the data center based on software defined networking (SDN). In order to improve InPs revenue, we consider the tradeoff of virtual data center (VDC) reliability requirement and the
Jianqiang Ma is a Ph.D student in the School of Public Administration, Beihang University. His main interesting research areas are resource management, cloud computing and big data.
References (16)
- et al.
Software defined networks: a survey
J Netw Comput Appl
(2016) - et al.
Joint flow routing-scheduling for energy efficient software defined data center networks
J Netw Comput Appl
(2016) - et al.
Resource allocation with multi factor node ranking in data center networks
Future Gen Comput Syst
(2014) - et al.
Data center network virtualization: a survey
IEEE Commun Surv Tut
(2013) - et al.
Greenhead virtual data center embedding across distributed infrastructures
IEEE Trans Cloud Comput
(2013) - et al.
Improving network management with software defined networking
IEEE Commun Mag
(2013) - et al.
Virtual network embedding; a survey
IEEE Commun Surv Tut
(2013) - et al.
On achieving high survivability in virtualized data centers
IEICE Trans Commun
(2014)
Cited by (7)
Implementing an intrusion detection and prevention system using Software-Defined Networking: Defending against ARP spoofing attacks and Blacklisted MAC Addresses
2021, Computers and Electrical EngineeringCitation Excerpt :This allows implementation of an efficient global configuration across its entirety, making overall network management easier. The centralised control plane, which consists of one or more controllers, improves network performance and enables more efficient network monitoring [2]. SDN-enabled applications send their requests to the controller, which in turn, directly communicates them to the data plane.
Application association and load balancing to enhance energy efficiency in heterogeneous wireless networks
2018, Computers and Electrical EngineeringCitation Excerpt :The load-balancing scheme was implemented at a control centre with concentrated computing capabilities to balance the loads of the RRHs and APs. Cloud radio access network (C-RAN) architecture with software-defined networking (SDN) [5] was adopted. The weights were calculated and used to play a cooperative game that allocated wireless network resources among the MDs to achieve energy efficiency [6].
Optimal network-aware virtual data center embedding
2023, International Journal of Network ManagementResearch on fog resource scheduling based on cloud-fog collaboration technology in the electric internet of things
2021, Recent Advances in Electrical and Electronic EngineeringA survey and taxonomy on virtual data center embedding
2019, Journal of SupercomputingResource availability–aware adaptive provisioning of virtual data center networks
2019, International Journal of Network Management
Jianqiang Ma is a Ph.D student in the School of Public Administration, Beihang University. His main interesting research areas are resource management, cloud computing and big data.
- ☆
Reviews processed and recommended for publication to the Editor-in-Chief by Guest Editor Dr. Z. Zheng.