Unsupervised reinforcement learning (RL) aims at pre-training models that can solve a wide range of downstream tasks in complex environments. Despite recent advancements, existing approaches suffer from several limitations: they may require running an RL process on each task to achieve a satisfactory performance, they may need access to datasets with good coverage or well-curated task-specific samples, or they may pre-train policies with unsupervised losses that are poorly correlated with the downstream tasks of interest. In this paper, we introduce FB-CPR, which regularizes unsupervised zero-shot RL based on the forward-backward (FB) method towards imitating trajectories from unlabeled behaviors. The resulting models learn useful policies imitating the behaviors in the dataset, while retaining zero-shot generalization capabilities. We demonstrate the effectiveness of FB-CPR in a challenging humanoid control problem. Training FB-CPR online with observation-only motion capture datasets, we obtain the first humanoid behavioral foundation model that can be prompted to solve a variety of whole-body tasks, including motion tracking, goal reaching, and reward optimization. The resulting model is capable of expressing human-like behaviors and it achieves competitive performance with task-specific methods while outperforming state-of-the-art unsupervised RL and model-based baselines.

Cloud service providers (CSPs) offer different Service Level Agreements (SLAs) to the cloud users. Cloud Service Brokers (CSBs) provide multiple sets of alternatives to the cloud users according to users requirements. Generally, a CSB considers the service commitments of CSPs rather than the actual quality of CSPs services. To overcome this issue, the broker should verify the service performances while recommending cloud services to the cloud users, using all available data. In this paper, we compare our two approaches to do so: a min-max-min decomposition based on Intuitionistic Fuzzy Logic (IFL) and a Performance Heat Map technique, to evaluate the performance of commercially available cloud providers. While the IFL technique provides simple, total order of the evaluated CSPs, Performance Heat Map provides transparent and explanatory, yet consistent evaluation of service performance of commercially available CSPs. The identified drawbacks of the IFL technique are: 1) It does not return the accurate performance evaluation over multiple decision alternatives due to highly influenced by critical feedback of the evaluators, 2) Overall ranking of the CSPs is not as expected according to the performance measurement. As a result, we recommend to use performance Heat Map for this problem.

The knowledge of service performance of cloud providers is essential for cloud service users to choose the cloud services that meet their requirements. Instantaneous performance readings are accessible, but prolonged observations provide more reliable information. However, due to technical complexities and costs of monitoring services, it may not be possible to access the service performance of cloud provider for longer time durations. The extended observation periods are also a necessity for prediction of future behavior of services. These predictions have very high value for decision making both for private and corporate cloud users, as the uncertainty about the future performance of purchased cloud services is an important risk factor. Predictions can be used by specialized entities, such as cloud service brokers (CSBs) to optimally recommend cloud services to the cloud users. In this paper, we address the challenge of prediction. To achieve this, the current service performance patterns of cloud providers are analyzed and future performance of cloud providers are predicted using to the observed service performance data. It is done using two automatic predicting approaches: ARIMA and ETS. Error measures of entire service performance prediction of cloud providers are evaluated against the actual performance of the cloud providers computed over a period of one month. Results obtained in the performance prediction show that the methodology is applicable for both short-term and long-term performance prediction.

This work presents an on-line, energy- and communication-aware scheduling strategy for SaaS applications in data centers. The applications are composed of various services and represented as workflows. Each workflow consists of tasks related to each other by precedence constraints and represented by Directed Acyclic Graphs (DAGs). The proposed scheduling strategy combines advantages of state-of-the-art workflow scheduling strategies with energy-aware independent task scheduling approaches. The process of scheduling consists of two phases. In the first phase, virtual deadlines of individual tasks are set in the central scheduler. These deadlines are determined using a novel strategy that favors tasks which are less dependent on other tasks. During the second phase, tasks are dynamically assigned to computing servers based on the current load of network links and servers in a data center. The proposed approach, called Minimum Dependencies Energy-efficient DAG (MinD+ED) scheduling, has been implemented in the GreenCloud simulator. It outperforms other approaches in terms of energy efficiency, while keeping a satisfiable level of tardiness.

Heterogeneous architectures have become more popular and widespread in the recent years with the growing popularity of general-purpose processing on graphics processing units, low-power systems on a chip, multi- and many-core architectures, asymmetric cores, coprocessors, and solid-state drives. The design and management of cloud computing data-centers must adapt to these changes while targeting objectives of improving system performance, energy efficiency and reliability. This paper presents HEROS, a novel load balancing algorithm for energy-efficient resource allocation in heterogeneous systems. HEROS takes into account the heterogeneity of a system during the decision-making process and uses a holistic representation of the system. As a result, servers that contain resources of multiple types (computing, memory, storage and networking) and have varying internal structures of their components can be utilized more efficiently.

Selecting the appropriate cloud services and cloudproviders according to the cloud users requirements is becoming a complex task, as the number of cloud providers increases. Cloud providers offer similar kinds of cloud services, but they are different in terms of price, quality of service, customer experience, and service delivery. The most challenging issue of the current cloud computing business is that cloud providers commit a certain Service Level Agreement (SLA), with cloud users, but there is little or no verification mechanisms which ensure that cloud providers are providing cloud services according to their commitment. In the current literature, there is a lack of an evaluation model which provides the real status of cloud providers for the cloud users. In this paper, an evaluation model is proposed, which verifies the quality of cloud services delivered for each service and provides the service status of the cloud providers. Finally, evaluation results obtained from cloud auditors are visualized in an ordered performance heat map, showing the cloud providers in a decreasing ordering of overall service quality. In this way, the proposed service quality evaluation model represents a visual recommender system for cloud service brokers and cloud users.

SUMMARY Virtualization is emerging as the prominent approach to mutualise the energy consumed by a single server running multiple virtual machines instances. The efficient utilisation of virtualized servers and/or computing resources requires understanding of the overheads in energy consumption and the throughput especially on high‐demanding high‐performance computing (HPC) platforms. In this paper, a novel holistic model for the power of virtualized computing nodes is proposed. Moreover, we create and validate instances of the proposed model using concrete measures taken during a benchmarking process that reflects an HPC usage, that is, HPC challenge, IOZone and Bonnie++, conducted using two different hardware configurations on Grid'5000 platform, based on Intel and Advanced Micro Devices (AMD) processors and three widespread virtualization frameworks, namely, Xen, Kernel‐based virtual machine and VMware ESXi. The proposed holistic model of machine power takes into account the impact of utilisation metrics of the machine's components, as well as the employed application, virtualization and hardware. The model is further derived using tools such as multiple linear regressions or neural networks that prove its elasticity, applicability and accuracy. The purpose of the model is to enable the estimation of energy consumption of virtualized platforms, aiming to make possible the optimization, scheduling or accounting in such systems or their simulation. Copyright © 2014 John Wiley & Sons, Ltd.

Management and optimization of cloud infrastructures combine multiple challenges. The optimization of data centers targets such objectives as performance, reliability, energy consumption, and security. To achieve these goals, multiple actions can be taken, for example, task and virtual machine allocation or infrastructure management. In this work we propose a model for representation of computing, memory, storage, and communication resources in cloud computing data centers. This model is relevant for the characterization of cloud applications, virtual machines, as well as physical servers. The performance evaluation and validation of the proposed model is carried out using the Green Cloud simulator. The obtained results show good agreement with the design objectives and confirm validity of the assumptions.

With a growing concern on the considerable energy consumed by HPC platforms and data centers, research efforts are targeting green approaches with higher energy efficiency. In particular, virtualization is emerging as the prominent approach to mutualize the energy consumed by a single server running multiple VMs instances. Even today, it remains unclear whether the overhead induced by virtualization and the corresponding hypervisor middleware suits an environment as high-demanding as an HPC platform. In this paper, we analyze from an HPC perspective the three most widespread virtualization frameworks, namely Xen, KVM, and VMware ESXi and compare them with a baseline environment running in native mode. We performed our experiments on the Grid'5000 platform by measuring the results of the reference HPL benchmark. Power measures were also performed in parallel to quantify the potential energy efficiency of the virtualized environments. In general, our study offers novel incentives toward in-house HPC platforms running without any virtualized frameworks.