Resource Utilization In Cloud Data Centers Research Articles

Jointly Optimizing Job Assignment and Resource Partitioning for Improving System Throughput in Cloud Datacenters

Colocating multiple jobs on the same server has been widely applied for improving resource utilization in cloud datacenters. However, the colocated jobs would contend for the shared resources, which could lead to significant performance degradation. An efficient approach for eliminating performance interference is to partition the shared resources among the colocated jobs. However, this makes the resource management in datacenters very challenging. In this paper, we propose JointOPT, the first resource management framework that optimizes job assignment and resource partitioning jointly for improving the throughput of cloud datacenters. JointOPT uses a local search based algorithm to find the near optimal job assignment configuration, and uses a deep reinforcement learning (DRL) based approach to dynamically partition the shared resources among the colocated jobs. In order to reduce the interaction overhead with real system, it leverages deep learning to estimate job performance without running them on real servers. We conduct extensive experiments to evaluate JointOPT and the results show that JointOPT significantly outperform the state-of-the-art baselines, with an advantage from 13.3% to 47.7%.

A container deployment strategy for server clusters with different resource types

AbstractThe method of deploying microservices based on container technology is widely used in cloud environments. This method can realize the rapid deployment of microservices and improve the resource utilization of cloud datacenters. However, resource allocation and deployment of container‐based microservices are key issues. With the continuous growth of computing‐ and storage‐intensive services, it is necessary to consider the deployment of microservices of different business types. This study establishes a multi‐objective optimization problem model with the similarity between containers and servers, load balance of clusters, and reliability of microservice execution as the optimization objectives. An improved artificial fish swarm algorithm is proposed for the container deployment of computing‐ and storage‐intensive microservices. The comprehensive experimental results show that, compared with the existing deployment strategies, the matching degree between the container and server, cluster load balance value, service execution reliability, and other performance parameters are improved while shortening the running time of the algorithm. In addition, under the constraint of load balancing, the resource utilization of the computing and storage server clusters is improved.

Classification-Based and Energy-Efficient Dynamic Task Scheduling Scheme for Virtualized Cloud Data Center

The size and number of cloud data centers (CDCs) have grown rapidly with the increasing popularity of cloud computing and high-performance computing. This has the unintended consequences of creating new challenges due to inefficient use of resources and high energy consumption. Hence, this necessitates the need to maximize resource utilization and ensure energy efficiency in CDCs. One viable approach to achieve energy efficiency and resource utilization in CDC is task scheduling. While several task scheduling approaches have been proposed in the literature, there appears to be a lack of classification-based merging concept for real-time tasks in these existing approaches. Thus, an energy-efficient dynamic scheduling scheme (EDS) of real-time tasks for virtualized CDC is presented in this paper. In the scheduling scheme, the heterogeneous tasks and virtual machines are first classified based on a historical scheduling record. Then, similar type of tasks are merged and scheduled to maximally utilize an operational state of the host. In addition, energy efficiencies and optimal operating frequencies of heterogeneous physical hosts are employed to attain energy preservation while creating and deleting the virtual machines. Experimental results show that, in comparison with existing techniques, EDS significantly improves overall scheduling performance, achieves a higher CDC resource utilization, increases task guarantee ratio, minimizes the mean response time, and reduces energy consumption.

Stochastic Resource Provisioning for Containerized Multi-Tier Web Services in Clouds

Under today's bursty web traffic, the fine-grained per-container control promises more efficient resource provisioning for web services and better resource utilization in cloud datacenters. In this paper, we present Two -stage S tochastic P rogramming R esource A llocator (2SPRA). It optimizes resource provisioning for containerized n-tier web services in accordance with fluctuations of incoming workload to accommodate predefined SLOs on response latency. In particular, 2SPRA is capable of minimizing resource over-provisioning by addressing dynamics of web traffic as workload uncertainty in a native stochastic optimization model. Using special-purpose OpenOpt optimization framework, we fully implement 2SPRA in Python and evaluate it against three other existing allocation schemes, in a Docker-based CoreOS Linux VMs on Amazon EC2. We generate workloads based on four real-world web traces of various traffic variations: AOL, WorldCup98, ClarkNet, and NASA. Our experimental results demonstrate that 2SPRA achieves the minimum resource over-provisioning outperforming other schemes. In particular, 2SPRA allocates only 6.16 percent more than application's actual demand on average and at most 7.75 percent in the worst case. It achieves 3x further reduction in total resources provisioned compared to other schemes delivering overall cost-savings of 53.6 percent on average and up to 66.8 percent. Furthermore, 2SPRA demonstrates consistency in its provisioning decisions and robust responsiveness against workload fluctuations.

Building Efficient Resource Management Systems in the Cloud: Opportunities and Challenges

Cloud computing is designed as computing as a utility. Customers rent computing resources in the cloud to complete their work. To ensure the quality of service (QoS) requirements defined by customers and guarantee the resource utilization in cloud datacenters, effective resource management systems should be considered. However, with more and more individuals and enterprises migrating their work into the cloud, workloads in the cloud become more and more heterogeneous. Meanwhile, resources are much more heterogeneous as cloud providers constantly scale or update the clusters with new generations of machines. Withal, workloads are dynamic with different resource demands during their execution. Besides, machines are re-engaged frequently after removal due to various reasons such as crushing and updating. The heterogeneity and dynamicity in both workloads and resources are huge barriers to using classic resource management systems. This paper will first introduce the status quo of cloud computing environment, and then give an overview of resource demand prediction and allocation policies. Finally, challenges are proposed to help build adaptive resource management systems.Keywords