Intelligent Virtual Machine Provisioning in Cloud Computing

IJCAI, pp. 1495-1502, 2020.

Cited by: 0|Bibtex|Views12|Links
EI
Keywords:
Best from Multiple Selectionslong short-term memoryUncertainty-Aware Heuristic Searchminimum vertex coverexpected improvementMore(20+)
Weibo:
We propose Uncertainty-Aware Heuristic Search, which is a novel approach for solving predictive VM provisioning

Abstract:

Virtual machine (VM) provisioning is a common and critical problem in cloud computing. In industrial cloud platforms, there are a huge number of VMs provisioned per day. Due to the complexity and resource constraints, it needs to be carefully optimized to make cloud platforms effectively utilize the resources. Moreover, in practice, provi...More

Code:

Data:

0
Introduction
  • Cloud computing has emerged as a new computing paradigm that offers a variety of services [Wang et al, 2015].
  • Virtual machine (VM) provisioning is a common and critical problem in cloud computing.
  • Due to its complicated resource constraints, the VM provisioning problem is computationally hard and urgently calls for effective solutions [Hbaieb et al, 2017].
  • In current industrial practice, when a customer requests for a VM, the VM is provisioned from scratch, which usually costs much time [Mao and Humphrey, 2012].
  • It is advisable to provision VMs ahead, and to serve requests with provisioned VMs
Highlights
  • Cloud computing has emerged as a new computing paradigm that offers a variety of services [Wang et al, 2015]
  • Most steps in Virtual machine provisioning are independent to a particular request, which can be done before request
  • We propose Uncertainty-Aware Heuristic Search (UAHS), which is a novel approach for solving predictive VM provisioning
  • We propose Uncertainty-Aware Heuristic Search, a novel approach for predictive VM provisioning
  • We have successfully applied our Uncertainty-Aware Heuristic Search approach to Pre-Provisioning Service in Microsoft Azure and significantly improved the performance of the Virtual machine provisioning on the cloud platform
  • Uncertainty-Aware Heuristic Search has been successfully applied in Microsoft Azure and brought practical benefits in real-world applications
Methods
  • To study the performance of UAHS, the authors conduct extensive experiments on two public datasets and an industrial dataset to compare UAHS against 8 state-of-the-art competitors.
  • In order to capture the weekly characteristics, the authors construct 42 instances by taking all sub-time series in length of 139 (i.e., 180 − 42 + 1) time stamps using the sliding window based time series analysis approach [Box et al, 2015], each of which has the demand on the last time stamp as label
  • Since both the Azure-2017 and Azure-2019 datasets lack the PM status, for each instance in the Azure-2017 and Azure-2019 datasets, the PM status is generated synthetically to simulate a practical setup
Results
  • Comparisons against state-of-the-art competitors.
  • The comparative results of UAHS and its 8 state-of-the-art competitors on all 3 datasets are presented in Table 1.
  • It is clear prob that UAHS stands out as the best method in terms of average utilization ratio and average run time.
  • On all 3 datasets UAHS achieves the utilization ratio more than 0.8, while the figures for its all competitors are less than 0.8.
  • When the authors focus on the metric of average run time, UAHS runs much faster than its all competitors.
  • The results in Table 1 indicate both the effectiveness and the efficiency of UAHS
Conclusion
  • Virtual machine (VM) provisioning is a common and critical problem and it is advisable to provision VMs ahead for customer experience.
  • The authors formulated the predictive VM provisioning (PreVMP) problem, and proposed a novel approach, dubbed Uncertainty-Aware Heuristics Search (UAHS), for solving the PreVMP problem.
  • UAHS models and utilizes the prediction uncertainty to conduct optimization.
  • UAHS leverages Bayesian optimization to interact the prediction component and the optimization component, to achieve performance improvement.
  • Extensive experiments on two public datasets and an industrial dataset show that UAHS performs much better than stateof-the-art competitors.
  • UAHS has been successfully applied in Microsoft Azure and brought practical benefits in real-world applications
Summary
  • Introduction:

    Cloud computing has emerged as a new computing paradigm that offers a variety of services [Wang et al, 2015].
  • Virtual machine (VM) provisioning is a common and critical problem in cloud computing.
  • Due to its complicated resource constraints, the VM provisioning problem is computationally hard and urgently calls for effective solutions [Hbaieb et al, 2017].
  • In current industrial practice, when a customer requests for a VM, the VM is provisioned from scratch, which usually costs much time [Mao and Humphrey, 2012].
  • It is advisable to provision VMs ahead, and to serve requests with provisioned VMs
  • Methods:

    To study the performance of UAHS, the authors conduct extensive experiments on two public datasets and an industrial dataset to compare UAHS against 8 state-of-the-art competitors.
  • In order to capture the weekly characteristics, the authors construct 42 instances by taking all sub-time series in length of 139 (i.e., 180 − 42 + 1) time stamps using the sliding window based time series analysis approach [Box et al, 2015], each of which has the demand on the last time stamp as label
  • Since both the Azure-2017 and Azure-2019 datasets lack the PM status, for each instance in the Azure-2017 and Azure-2019 datasets, the PM status is generated synthetically to simulate a practical setup
  • Results:

    Comparisons against state-of-the-art competitors.
  • The comparative results of UAHS and its 8 state-of-the-art competitors on all 3 datasets are presented in Table 1.
  • It is clear prob that UAHS stands out as the best method in terms of average utilization ratio and average run time.
  • On all 3 datasets UAHS achieves the utilization ratio more than 0.8, while the figures for its all competitors are less than 0.8.
  • When the authors focus on the metric of average run time, UAHS runs much faster than its all competitors.
  • The results in Table 1 indicate both the effectiveness and the efficiency of UAHS
  • Conclusion:

    Virtual machine (VM) provisioning is a common and critical problem and it is advisable to provision VMs ahead for customer experience.
  • The authors formulated the predictive VM provisioning (PreVMP) problem, and proposed a novel approach, dubbed Uncertainty-Aware Heuristics Search (UAHS), for solving the PreVMP problem.
  • UAHS models and utilizes the prediction uncertainty to conduct optimization.
  • UAHS leverages Bayesian optimization to interact the prediction component and the optimization component, to achieve performance improvement.
  • Extensive experiments on two public datasets and an industrial dataset show that UAHS performs much better than stateof-the-art competitors.
  • UAHS has been successfully applied in Microsoft Azure and brought practical benefits in real-world applications
Tables
  • Table1: Results of UAHS, UAHS-alt1, UAHS-alt2 and their competitors on all datasets
  • Table2: Results of UAHS with different hyper-parameter settings of max _iter on all datasets
  • Table3: Results of UAHS with different hyper-parameter settings of prob on all datasets
Download tables as Excel
Related work
  • The deterministic VM provisioning (VMP) problem aims to find a feasible provisioning plan that can optimize the allocations of VMs to PMs under resource constraints. A number of methods have been proposed to solve deterministic VMP [Hbaieb et al, 2017; Zhao et al, 2018], and the Ant Colony Optimization (ACO) algorithm [Zhao et al, 2018] exhibits state-of-the-art performance. However, these methods are only applicable when the real VM demands are known.

    For PreVMP, cloud platforms need to provision VMs according to the upcoming demands, which unfortunately are previously unknown. Hence, the PreVMP problem can be treated as a problem of prediction with optimization (Prediction+Optimization) [Wilder et al, 2019]. A straightforward two-stage method first predicts unknown parameters and then directly conducts optimization based on the predicted results. An improved two-stage method called Semidirect [Demirovicet al., 2019a] is proposed by modifying the loss function according to the characteristics of the optimization problem. Unfortunately, such two-stage methods assume that the predicted results are accurate, but in practice the prediction errors are inevitable [Wilder et al, 2019].
Reference
  • [Box et al., 2015] George E. P. Box, Gwilym M. Jenkins, Gregory C. Reinsel, and Greta M. Ljung. Time Series Analysis: Forecasting and Control. Wiley, 2015.
    Google ScholarFindings
  • [Cai et al., 2013] Shaowei Cai, Kaile Su, Chuan Luo, and Abdul Sattar. NuMVC: An efficient local search algorithm for minimum vertex cover. Journal of Artificial Intelligence Research, 46:687–716, 2013.
    Google ScholarLocate open access versionFindings
  • [Cai et al., 2016] Shaowei Cai, Chuan Luo, Jinkun Lin, and Kaile Su. New local search methods for partial MaxSAT. Artificial Intelligence, 240:1–18, 2016.
    Google ScholarLocate open access versionFindings
  • [Cai, 2015] Shaowei Cai. Balance between complexity and quality: Local search for minimum vertex cover in massive graphs. In Proceedings of IJCAI 2015, pages 747–753, 2015.
    Google ScholarLocate open access versionFindings
  • [Cortez et al., 2017] Eli Cortez, Anand Bonde, Alexandre Muzio, Mark Russinovich, Marcus Fontoura, and Ricardo Bianchini. Resource central: Understanding and predicting workloads for improved resource management in large cloud platforms. In Proceedings of SOSP 2017, pages 153–167, 2017.
    Google ScholarLocate open access versionFindings
  • [Demirovicet al., 2019a] Emir Demirovic, Peter J. Stuckey, James Bailey, Jeffrey Chan, Chris Leckie, Kotagiri Ramamohanarao, and Tias Guns. An investigation into prediction + optimisation for the knapsack problem. In Proceedings of CPAIOR 2019, pages 241–257, 2019.
    Google ScholarLocate open access versionFindings
  • [Demirovicet al., 2019b] Emir Demirovic, Peter J. Stuckey, James Bailey, Jeffrey Chan, Christopher Leckie, Kotagiri Ramamohanarao, and Tias Guns. Predict+optimise with ranking objectives: Exhaustively learning linear functions. In Proceedings of IJCAI 2019, pages 1078–1085, 2019.
    Google ScholarLocate open access versionFindings
  • [Demirovicet al., 2020] Emir Demirovic, Peter J. Stuckey, James Bailey, Jeffrey Chan, Christopher Leckie, Kotagiri Ramamohanarao, and Tias Guns. Dynamic programming for predict+optimise. In Proceedings of AAAI 2020, 2020.
    Google ScholarLocate open access versionFindings
  • [Durbin and Koopman, 2012] James Durbin and Siem Jan Koopman. Time Series Analysis by State Space Methods. Oxford University Press, 2012.
    Google ScholarFindings
  • [Elmachtoub and Grigas, 2017] Adam N. Elmachtoub and Paul Grigas. Smart “Predict, then optimize”. CoRR, abs/1710.08005, 2017.
    Findings
  • [Hbaieb et al., 2017] Ameni Hbaieb, Mahdi Khemakhem, and Maher Ben Jemaa. Using decomposition and local search to solve large-scale virtual machine placement problems with disk anti-colocation constraints. In Proceedings of AICCSA 2017, pages 688–695, 2017.
    Google ScholarLocate open access versionFindings
  • [Hoos and Stützle, 2004] Holger H. Hoos and Thomas Stützle. Stochastic Local Search: Foundations & Applications. Elsevier / Morgan Kaufmann, 2004.
    Google ScholarFindings
  • [Hutter et al., 2011] Frank Hutter, Holger H. Hoos, and Kevin Leyton-Brown. Sequential model-based optimization for general algorithm configuration. In Proceedings of LION 2011, pages 507–523, 2011.
    Google ScholarLocate open access versionFindings
  • [Hyndman and Athanasopoulos, 2013] Rob J. Hyndman and George Athanasopoulos. Forecasting: Principles and Practice. OTexts, 2013.
    Google ScholarLocate open access versionFindings
  • [Jones et al., 1998] Donald R. Jones, Matthias Schonlau, and William J. Welch. Efficient global optimization of expensive black-box functions. Journal of Global Optimization, 13(4):455–492, 1998.
    Google ScholarLocate open access versionFindings
  • [Luo et al., 2015] Chuan Luo, Shaowei Cai, Wei Wu, Zhong Jie, and Kaile Su. CCLS: An efficient local search algorithm for weighted maximum satisfiability. IEEE Transactions on Computers, 64(7):1830–1843, 2015.
    Google ScholarLocate open access versionFindings
  • [Luo et al., 2017] Chuan Luo, Shaowei Cai, Kaile Su, and Wenxuan Huang. CCEHC: An efficient local search algorithm for weighted partial maximum satisfiability. Artificial Intelligence, 243:26–44, 2017.
    Google ScholarLocate open access versionFindings
  • [Luo et al., 2019] Chuan Luo, Holger H. Hoos, Shaowei Cai, Qingwei Lin, Hongyu Zhang, and Dongmei Zhang. Local search with efficient automatic configuration for minimum vertex cover. In Proceedings of IJCAI 2019, pages 1297– 1304, 2019.
    Google ScholarLocate open access versionFindings
  • [Mandi et al., 2020] Jaynta Mandi, Emir Demirovic, Peter. J Stuckey, and Tias Guns. Smart predict-and-optimize for hard combinatorial optimization problems. In Proceedings of AAAI 2020, 2020.
    Google ScholarLocate open access versionFindings
  • [Mao and Humphrey, 2012] Ming Mao and Marty Humphrey. A performance study on the VM startup time in the cloud. In Proceedings of IEEE CLOUD 2012, pages 423–430, 2012.
    Google ScholarLocate open access versionFindings
  • [McLachlan and Peel, 2000] Geoffrey McLachlan and David Peel. Finite Mixture Models. Wiley, 2000.
    Google ScholarFindings
  • [Mockus, 1989] Jonas Mockus. Bayesian Approach to Global Optimization: Theory and Applications. Kluwer Academic Publishers, 1989.
    Google ScholarFindings
  • [Shahriari et al., 2016] Bobak Shahriari, Kevin Swersky, Ziyu Wang, Ryan P. Adams, and Nando de Freitas. Taking the human out of the loop: A review of Bayesian optimization. Proceedings of the IEEE, 104(1):148–175, 2016.
    Google ScholarLocate open access versionFindings
  • [Taylor and Letham, 2018] Sean J. Taylor and Benjamin Letham. Forecasting at scale. The American Statistician, 72(1):37–45, 2018.
    Google ScholarLocate open access versionFindings
  • [Verbeek et al., 2003] Jakob J. Verbeek, Nikos A. Vlassis, and Ben J. A. Kröse. Efficient greedy learning of Gaussian mixture models. Neural Computation, 15(2):469– 485, 2003.
    Google ScholarLocate open access versionFindings
  • [Wang et al., 2015] Changjun Wang, Weidong Ma, Tao Qin, Xujin Chen, Xiaodong Hu, and Tie-Yan Liu. Selling reserved instances in cloud computing. In Proceedings of IJCAI 2015, pages 224–231, 2015.
    Google ScholarLocate open access versionFindings
  • [Wilder et al., 2019] Bryan Wilder, Bistra Dilkina, and Milind Tambe. Melding the data-decisions pipeline: Decision-focused learning for combinatorial optimization. In Proceedings of AAAI 2019, pages 1658–1665, 2019.
    Google ScholarLocate open access versionFindings
  • [Zhang et al., 2014] Zhaoning Zhang, Ziyang Li, Kui Wu, Dongsheng Li, Huiba Li, Yuxing Peng, and Xicheng Lu. VMThunder: Fast provisioning of large-scale virtual machine clusters. IEEE Transactions on Parallel and Distributed Systems, 25(12):3328–3338, 2014.
    Google ScholarLocate open access versionFindings
  • [Zhao et al., 2018] Hui Zhao, Jing Wang, Feng Liu, Quan Wang, Weizhan Zhang, and Qinghua Zheng. Power-aware and performance-guaranteed virtual machine placement in the cloud. IEEE Transactions on Parallel and Distributed Systems, 29(6):1385–1400, 2018.
    Google ScholarLocate open access versionFindings
Your rating :
0

 

Tags
Comments