Energy-aware virtual machine allocation for cloud with resource reservation☆
Introduction
Based on the virtualization of computing resources, cloud computing enables on-demand service provision in a pay-as-you-go manner, which makes the upgrades and maintenance of both software and hardware easier than before (Buyya et al., 2009). Therefore, instead of incurring high upfront costs in building their own private platforms, more and more enterprises and communities choose cloud computing platforms to deploy their commercial and scientific applications. However, the proliferation of cloud computing requires the establishment of large-scale data centers that contain thousands of computing nodes, which in turn results in excessive energy consumption and negative environmental impacts (Wajid et al., 2016).
Although the execution of more cloud services needs more energy, the energy consumption in cloud data centers is mainly due to the low utilization of computing resources. It has been estimated that the average resource utilization in most data centers is lower than 30% (Barroso et al., 2013) and the energy consumed by idle nodes is more than 70% of peak energy (Fan et al., 2007a). In other words, most energy is wasted for doing nothing. The rising energy consumption increases the ownership cost and reduces the return on cloud infrastructure investments. Therefore, how to fully explore the capacity of data centers to achieve high energy efficiency is becoming a major concern of cloud service providers.
To efficiently manage cloud resources and reduce the unit price, an increasing number of cloud providers support the resource reservation option, which allows tenants to customize their cloud service requests. For example, Amazon Elastic Compute Cloud (EC2) (Cloud, 2016) has the option Reserved Instances Pricing which allows cloud capacity reservations within specific time windows (i.e., a fraction of a day or a week). Such a kind of computing mode is also supported by Microsoft Azure automation, Alibaba Cloud Batch Compute, and other mainstream cloud platforms. Fig. 1 shows how virtual machine requests are handled in the workflow of reservation-based cloud computing. Initially, cloud tenants submit their virtual machine reservation requests with specific requirements (e.g., required type of resources, time windows) to an IaaS (Infrastructure as a Service) provider. For example, if one tenant wants to conduct 3D rendering using the reservation-based cloud, he/she needs to provide the start time, CPU and GPU requirements and the longest rendering time to the IaaS provider. After receiving such a batch of VM reservations, the IaaS provider needs to figure out a new Virtual Machine (VM) to Physical Machine (PM) mapping based on available resources. From the perspective of IaaS providers, such a mapping should achieve maximum profit and meet all the tenants’ requirements. Meanwhile, the newly incorporated VMs should not degrade the performance of other existing VMs in service. If there are not enough physical resources that can accommodate all the VMs, i.e, the Service Level Agreement (SLA) of some VM requests cannot be ensured, the unsatisfied requests will be rejected by their IaaS providers. When the reservation is confirmed, at the beginning of the next VM reservation cycle, all the accepted VMs will be dispatched to their designated PMs.
The VM-to-PM mapping plays an important role in determining the utilization of data centers (Pietri, Sakellariou, 2016, Chen, Huang, Fu, Liu, He). To achieve more profits in the very competitive cloud computing market, IaaS providers need to explore efficient VM-to-PM mappings that can achieve the highest energy efficiency while consolidating as much VM workload as possible. Since the VM allocation problem is an NP-hard problem, various energy-aware VM mapping approaches have been investigated (Kim, Beloglazov, Buyya, 2011, Calheiros, Buyya, 2014, Hwang, Pedram, 2013). However, most of them are based on the trade-off between energy consumption and system performance assuming that there are unlimited physical resources. This is not suitable for reservation-based cloud computing. In order to fully utilize the PMs and facilitate the management of reserved VMs, IaaS providers usually offer VM reservation on a limited number of exclusive PMs without mixing reserved and non-reserved VMs together. Consequently, VM rejection should be considered during the mapping generation. Moreover, due to the increasing new VM requests, the reservation cycle time is shortened drastically (e.g., from one week to one day). Accordingly, the update of reservation plans is becoming much more frequent.
In order to save operating cost and quickly respond to the new VM requests, IaaS providers should consider the following questions during the derivation of mapping solutions: (i) how to quickly achieve a VM-to-PM mapping which has the highest energy-efficiency? and (ii) how to incorporate more new VM requests to achieve the highest PM utilization without violating tenants’ requirements? To address these two questions, this paper makes the following three major contributions:
- •
We propose a novel fitness function based on our defined term instruction-energy ratio, which can effectively reduce the overall energy consumption and maximize the resource utilization of reservation-based cloud data centers.
- •
Based on the evolutionary algorithm, we introduce a comprehensive VM allocation approach that can efficiently converge to a VM-to-PM mapping with best possible energy efficiency.
- •
To further accelerate the exploration of optimal VM allocation solutions, we develop an efficient simulation engine and integrate it into the cloud simulator CloudSim (Calheiros et al., 2011), which can drastically reduce the evaluation time of VM allocation solutions in each evolutionary iteration.
The rest of the paper is organized as follows. Section 2 presents related work on energy-aware VM allocation. After an introduction to the modeling and problem definition of energy-aware reservation-based VM allocation in Section 3, Section 4 details our evolutionary VM allocation approach. Section 5 evaluates our approach using both simulation-based and real-world examples. Finally, Section 6 concludes the paper.
Section snippets
Related work
Due to the proliferation of cloud computing and escalation of energy consumption and operational cost, sustainable computing has become a major concern of cloud service providers. To reduce the energy consumption, various energy-aware approaches have been studied (Pietri, Sakellariou, 2016, Berl, Gelenbe, Girolamo, Giuliani, Meer, Dang, Pentikousis, 2010, Beloglazov, Abawajy, Buyya, 2012, Goudarzi, Pedram, 2016, Paya, Marinescu, 2017). For example, by formulating the VM consolidation problem as
System models and problem definition
We aim to maximize the energy efficiency (i.e., profit) of cloud data centers while minimizing the overall energy consumption with reserved VM requests. This section models the components of cloud data centers including PMs and tenant requests, and defines the problem that we are trying to solve.
Our evolutionary approach
Aiming at maximizing energy efficiency and serving more VMs in reservation-based cloud data centers, we detail our evolutionary approach with an illustrative example. To reduce the evaluation time of VM-to-PM mappings, we developed a new simulation engine on top of CloudSim, which can be used to drastically reduce the evaluation time of VM-to-PM mappings.
Performance evaluation
To validate the performance of our approach, we conducted experiments on both the simulation-based platform CloudSim (Calheiros et al., 2011) and a real cloud environment. We compared our approach with two well-known baseline methods (FF and MBFD Beloglazov et al., 2012) from the perspectives of overall profit, energy consumption, instruction-energy ratio, and request acceptance ratio.
Conclusions and future work
Service reservation is gaining popularity in cloud computing, since it can not only facilitate the management of increasing tenants’ requests, but also achieve lower price compared with traditional pay-as-you-go cloud services. However, when more and more virtual machines are deployed in data centers, the utilization rate of host PMs is becoming much more challenging to control. To address this problem, we proposed an evolutionary approach that can effectively allocate and consolidate VMs among
Xinqian Zhang received the B.E. degree from the Computer Science and Software Engineering Institute of East China Normal University in 2015. He is currently a Ph.D. student in the Computer Science and Software Engineering Institute, East China Normal University, Shanghai, China. His research interests are in the area of cloud computing, parallel and distributed systems, design automation of parallel computing systems, and software engineering.
References (34)
- et al.
Energy-aware resource allocation heuristics for efficient management of data centers for cloud computing
Future Gener. Comput. Syst.
(2012) - et al.
Cloud computing and emerging IT platforms: vision, hype, and reality for delivering computing as the 5th utility
Future Gener. Comput. Syst.
(2009) - et al.
VM Consolidation: a real case based on openstack cloud
Future Gener. Comput. Syst.
(2014) - et al.
Hierarchical SLA-driven resource management for peak power-aware and energy-efficient operation of a cloud datacenter
IEEE Trans. Cloud Comput.
(2016) - et al.
On achieving energy efficiency and reducing CO2 footprint in cloud computing
IEEE Trans. Cloud Comput.
(2016) - Amazon Elastic Compute Cloud, 2016. Amazon elastic compute cloud reserved instances pricing....
- et al.
The Datacenter as a Computer: An Introduction to the Design of Warehouse-Scale Machines
(2013) - et al.
Optimal online deterministic algorithms and adaptive heuristics for energy and performance efficient dynamic consolidation of virtual machines in cloud data centers
Concur. Comput Pract Exp.
(2012) - et al.
Energy-efficient cloud computing
Comput. J.
(2010) - et al.
Energy-efficient scheduling of urgent bag-of-tasks applications in clouds through DVFS
Proceedings of International Conference on Cloud Computing Technology and Science
(2014)
Cloudsim: a toolkit for modeling and simulation of cloud computing environments and evaluation of resource provisioning algorithms
Softw. Pract. Exper.
Energy-efficient virtual machines scheduling in multi-tenant data centers
IEEE Trans. Cloud Comput.
Power provisioning for a warehouse-sized computer
Proceedings of International Symposium on Computer Architecture (ISCA)
Power Provisioning for a warehouse-sized computer
Proceedings of International Symposium on Computer Architecture (ISCA)
Kraken: online and elastic resource reservations for multi-tenant datacenters
Proceedings of International Conference on Computer Communications (INFOCOM)
A multi-objective ant colony system algorithm for virtual machine placement in cloud computing
J. Comput. Syst. Sci.
Cited by (0)
Xinqian Zhang received the B.E. degree from the Computer Science and Software Engineering Institute of East China Normal University in 2015. He is currently a Ph.D. student in the Computer Science and Software Engineering Institute, East China Normal University, Shanghai, China. His research interests are in the area of cloud computing, parallel and distributed systems, design automation of parallel computing systems, and software engineering.
Tingming Wu received the B.E. degree from the Software Engineering Institute, East China Normal University, Shanghai, China, in 2015. He is currently a master student in the Institute of Computer Science and Software Engineering, East China Normal University. His research interests are in the area of cloud computing, parallel and distributed systems, design automation of embedded systems, and software engineering.
Mingsong Chen received the B.S. and M.E. degrees from Department of Computer Science and Technology, Nanjing University, Nanjing, China, in 2003 and 2006 respectively, and the Ph.D. degree in Computer Engineering from the University of Florida, Gainesville, in 2010. He is currently a Professor with the Computer Science and Software Engineering Institute at East China Normal University. His research interests are in the area of design automation of cyber-physical systems, cloud computing, parallel and distributed systems, and formal verification techniques. He is an Associate Editor of IET Computers & Digital Techniques, and Journal of Circuits, Systems and Computers.
Tongquan Wei received his Ph.D. degree in Electrical Engineering from Michigan Technological University in 2009. He is currently an Associate Professor in the Department of Computer Science and Technology at the East China Normal University. His research interests are in the areas of green and reliable embedded computing, cyber-physical systems, parallel and distributed systems, and cloud computing. He serves as a Regional Editor for Journal of Circuits, Systems, and Computers since 2012. He also served as Guest Editors for several special sections of IEEE TII and ACM TECS.
Junlong Zhou received his Ph.D. degree in Computer Science from East China Normal University in 2017. He is currently an Assistant Professor with the Department of Computer Science and Technology, Nanjing University of Science and Technology, Nanjing, China. He was a Research Visitor with the University of Notre Dame, Notre Dame, IN, USA. His current research interests include real-time embedded systems, cloud computing, and cyber-physical systems. Dr. Zhou has been an Associate Editor for the Journal of Circuits, Systems, and Computers since 2017, and a Guest Editor for the ACM Transactions on Cyber-Physical Systems.
Shiyan Hu received his Ph.D. in Computer Engineering from Texas A&M University in 2008. He is an Associate Professor at Michigan Tech, and he was a Visiting Associate Professor at Stanford University from 2015 to 2016. His research interests include Cyber-Physical Systems (CPS), CPS Security, Data Analytics, and Computer-Aided Design of VLSI Circuits, where he has published more than 100 refereed papers. He is an ACM Distinguished Speaker, an IEEE Systems Council Distinguished Lecturer, an IEEE Computer Society Distinguished Visitor, and a recipient of National Science Foundation (NSF) CAREER Award. Prof. Hu is the Chair for IEEE Technical Committee on Cyber-Physical Systems. He is the Editor-In-Chief of IET Cyber-Physical Systems: Theory&Applications. He is an Associate Editor for IEEE Transactions on Computer-Aided Design, IEEE Transactions on Industrial Informatics, and IEEE Transactions on Circuits and Systems. He has held chair positions in numerous IEEE/ACM conferences. He is a Fellow of IET and a senior member of the IEEE.
Rajkumar Buyya is a Redmond Barry Distinguished Professor and Director of the Cloud Computing and Distributed Systems (CLOUDS) Laboratory at the University of Melbourne, Australia. He served as a Future Fellow of the Australian Research Council during 2012–2016. He has authored over 625 publications and seven text books. He is one of the highly cited authors in computer science and software engineering worldwide. Dr. Buyya has led the establishment and development of key community activities, including serving as foundation Chair of the IEEE Technical Committee on Scalable Computing and five IEEE/ACM conferences. He served as the founding Editor-in-Chief of the IEEE Transactions on Cloud Computing. He is currently serving as Co-Editor-in-Chief of Journal of Software: Practice and Experience. Dr. Buyya is recognized as a “2016 Web of Science Highly Cited Researcher” by Thomson Reuters, a Scopus Researcher of the year 2017 with Excellence in Innovative Research Award by Elsevier, and a Fellow of IEEE for his outstanding contributions to Cloud computing.
- ☆
This work was supported by Natural Science Foundation of China (Grant Nos. 61872147 and 61802185), the Fundamental Research Funds for the Central Universities, Shanghai Municipal Natural Science Foundation (Grant No. 16ZR1409000), Natural Science Foundation of Jiangsu Province under Grant BK20180470, and China HGJ Project under Grant 2017ZX01038102-002.