Intelligent Virtual Machine Provisioning in Cloud Computing

Abstract

<jats:p>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, provisioning a VM from scratch requires fairly long time, which would degrade the customer experience. Hence, it is advisable to provision VMs ahead for upcoming demands. In this work, we formulate the practical scenario as the predictive VM provisioning (PreVMP) problem, where upcoming demands are unknown and need to be predicted in advance, and then the VM provisioning plan is optimized based on the predicted demands. Further, we propose Uncertainty-Aware Heuristic Search (UAHS) for solving the PreVMP problem. UAHS first models the prediction uncertainty, and then utilizes the prediction uncertainty in optimization. Moreover, UAHS leverages Bayesian optimization to interact prediction and optimization to improve its practical performance. Extensive experiments show that UAHS performs much better than state-of-the-art competitors on two public datasets and an industrial dataset. UAHS has been successfully applied in Microsoft Azure and brought practical benefits in real-world applications.</jats:p>

MetaNet: Automated Dynamic Selection of Scheduling Policies in Cloud Environments

“…Recent work in scheduling for cloud computing environments has demonstrated that AI-based solutions are not only faster, but can also scale efficiently compared to traditional heuristic and classical optimization techniques [5]- [8], [10]. Most contemporary dynamic resource management methods decouple the decision-making problem into two stages: QoS prediction and decision optimization [18]. This is commonly referred to as the predict+optimize framework in literature and is agnostic to the decision type [7].…”

“…Several methods have been proposed that leverage a forecasting model. For instance, a class of methods utilizes regression models such as Linear Regression (LR) [19] or Gaussian Process Regression [18], [20]. Others utilize auto-regressive models such as AutoARIMA [21] based forecasting.…”

A Surrogate Objective Framework for Prediction+Optimization with Soft Constraints

“…This set of optimization problems have the objective to match some expected quantities by penalizing excess and deficiency with probably different weights. Such formulation represents widespread resource provisioning problems, e.g., power [4] and cloud resources [1], where we minimize the cost of under-provisioning and over-provisioning against demands. Formally with α 1 , α 2 > 0, we have:…”

“…Mathematical optimization (a.k.a. mathematical programming), e.g., linear and quadratic programming, has been widely applied in decision-making processes, such as resource scheduling [1], goods production planning [2], portfolio optimization [3], and power scheduling [4]. In practice, problem parameters (e.g., goods demands, and equity returns) are often contextual and predicted by models with observed features (e.g., history time series).…”

Divide and Learn: A Divide and Conquer Approach for Predict+Optimize

“…Mandi et al (2020) also show that SPO+ loss can be used as a surrogate loss for relaxations of combinatorial problems and achieve performance improvements. Luo et al (2020) propose a specialised framework to optimize virtual machine provisioning. Black-box end to end frameworks are also used to differentiate and learn combinatorial problems (Bello et al 2016), (Li, Chen, and Koltun 2018), (Niculae et al 2018).…”