Distributed Data Centers Research Articles

Heating Homes with Servers: Workload Scheduling for Heat Reuse in Distributed Data Centers

Data centers consume lots of energy to execute their computational workload and generate heat that is mostly wasted. In this paper, we address this problem by considering heat reuse in the case of a distributed data center that features IT equipment (i.e., servers) installed in residential homes to be used as a primary source of heat. We propose a workload scheduling solution for distributed data centers based on a constraint satisfaction model to optimally allocate workload on servers to reach and maintain the desired home temperature setpoint by reusing residual heat. We have defined two models to correlate the heat demand with the amount of workload to be executed by the servers: a mathematical model derived from thermodynamic laws calibrated with monitored data and a machine learning model able to predict the amount of workload to be executed by a server to reach a desired ambient temperature setpoint. The proposed solution was validated using the monitored data of an operational distributed data center. The server heat and power demand mathematical model achieve a correlation accuracy of 11.98% while in the case of machine learning models, the best correlation accuracy of 4.74% is obtained for a Gradient Boosting Regressor algorithm. Also, our solution manages to distribute the workload so that the temperature setpoint is met in a reasonable time, while the server power demand is accurately following the heat demand.

Cost optimization of distributed data centers via computing workload distribution for next generation network systems

Recently, the research on sixth generation (6G) network systems has started as the deployments of fifth generation (5G) network systems go with a swing. One research aspect is the operation cost of 6G data centers because the energy consumption of the data centers has become a major portion of data centers’ operation cost as the request on 6G services gradually increases. Optimizing the energy consumption for lowering operating cost of data centers gains a lot of attention by the data center operators and becomes a hot issue of the current and future data center’s research. Most of the efforts in related studies have been made to minimize the operation cost from cooling, servers, and electricity aspects. However, most of the existing research studies address above operation cost issue in a separate manner which leads to inferior optimization results. Effective joint solutions are eagerly sought to optimize the operation cost of data centers. In this paper, a scheme of operation cost optimization is proposed for distributed data centers. Under the guide of the proposed scheme, electricity cost is optimized by taking into account of the multiple input single output (MISO) server power model, electricity price, and cooling efficiency. The workload distribution affecting the electricity cost as a major factor is studied and the optimal value of allocated workload is found to reduce the electricity cost. First, the whole system model is introduced. Next, based on the system model, the problem of cost optimization via workload dispatching is constructed without compromising service delay guarantee in service level agreement (SLA). Moreover, in order to conveniently solve the optimization problem, an algorithm is presented. Evaluation results demonstrate that the proposed scheme outperforms the existing scheme in terms of the total electricity cost reduction.

Explore unlabeled big data learning to online failure prediction in safety-aware cloud environment

Proactive fault management is an important problem in many areas of data management, including cloud computing, big data, vision, machine learning and especially for the cross-domain research of distributed computing and AI (Artificial Intelligence). Unfortunately, most real-world online failure prediction is facing the problem that the used data are difficult to label although the failure prediction should be a supervised learning problem. We observe that, in many cases, the large-scale unlabeled data can be classified through feature extraction and clustering for available prediction, and thus ideas from their combination can be brought to bear. Based on this, we have proposed an online failure prediction framework approach UDFP (Unlabeled Data based online Failure Prediction). It introduces the clustering analysis method based on the combination of the KNN (k-nearest neighbor) and the modularity idea to achieve prediction modeling. It is shown analytically that UDFP can mitigate a supervised learning problem for failure prediction in our situation to some extent. Experimental results demonstrate that UDFP, as a framework approach, has avoided the manual tagging workload and the huge difficulties, improved the predictive accuracy, and reduced cost of data management in safety-aware distributed cloud data centers while enhancing fault-tolerant capabilities and robustness.

Distributed clustering using multi-tier hierarchical overlay super-peer peer-to-peer network architecture for efficient customer segmentation

Customer segmentation divides customers into groups of consumers with similar patterns or characteristics relevant to marketing, especially with respect to target advertising. These patterns or characteristics could involve age, gender, interests, and spending habits. A customer segmentation model provides decision-makers with an effective allocation of marketing resources and maximization of cross-selling and up-selling opportunities. Clustering analysis is a key approach to understand business participants and obtain customers’ segments with similar demographics, behaviors, or trends. In e-business, online trade organizations store customer data in distributed data centers to limit the infrastructure or comply with the company norms. Incrementally, they need to update customer’s information to align the business offers with the contemporary trend. It is costly and exhaustive to collect the e-customer data at one location to exert centralized clustering. Therefore, distributed clustering is best suitable to categorize and segment customer data from inherently distributed resources. Currently, distributed architectures suffer from communication overhead or inaccurate global solutions. In this paper, a novel multi-layer hierarchical super-peer P2P (MT-SP2P) network architecture is proposed. The proposed MT-SP2P architecture provides a solution to enhance the speed of a distributed clustering problem while maintaining clustering quality. A novel distributed clustering algorithm is also proposed using the MT-SP2P architecture to improve the clustering speed without compromising the clustering quality. Customer segmentation is also seen as a managerial concept. Our distributed segmentation model and architecture help enterprises increase profits and improve customer service levels through effective and scalable customer segmentation. Computational results and managerial insights are discussed. We found from experimental results on different real customer data with various configurations and sizes. The proposed architecture and distributed clustering algorithm have improved the clustering speed by more than 90% compared to the centralized approaches. The findings show that the proposed model provided better insights and managerial implications concerning the chosen clustering techniques and distributed customer segments.

An Instance Reservation Framework for Cost Effective Services in Geo-Distributed Data Centers

Infrastructure-as-a-Service clouds in geo-distributed data centers offer various pricing options, including on-demand and reserved instances, which provide an elastic and cost-effective infrastructure to support High Performance Computing (HPC) applications. In this paper, we propose an instance reservation based cloud service framework, modeling the cost-minimizing reservation decision issue as an NP-hard integer programming problem for distributed data centers. To ease its computation complexity, two algorithms are proposed to minimize the HPC service cost with the worst-case performance guarantees: an offline heuristic-greedy algorithm, and a rolling-horizon based online algorithm when only short-term demand prediction is available. Facing fluctuating demands, instance reservation in a single data center may incur the highly underutilized capacity. To address this issue for further cost reduction, we extend the scheme with a novel cloud broker federation based resource sharing mechanism, reallocating already reserved but unused instances to computation-intensive and short-lived tasks for continuous execution without interruption. Extensive evaluations driven by large-scale trace-based datasets demonstrate that the proposed mechanism can effectively handle large volumes of service requests, saving considerable service costs with higher reservation resource utilization.

GreenCloudNet++: Simulation framework for energy efficient and secure, green job scheduling in geographically distributed data centers

AbstractGeographically distributed data centers are used as backbone infrastructure to meet rapidly increasing service demands of computations and data storage in cloud computing. This increase results in high energy consumption, increased operational expenditures, and high carbon footprint which are becoming points of great concern for service providers. In this research work, we present a simulation framework named GreenCloudNet++ for simulation and evaluation of energy‐efficient, green‐aware, and secure job scheduling mechanisms for geographically distributed data centers. GreenCloudNet++ has been developed as an extension to CloudNetSim++, which is designed to simulate distributed data center architectures connected with high speed networks. The functionality of CloudNetSim++ is extended by adding hierarchical job scheduling mechanism, hierarchical statistics collection mechanism and integration mechanism for green energy. Proposed model considers availability of green energy at each data center and maximizes its utilization. It considers the amount of underutilized computational resources at individual data centers while assigning jobs to the data centers, which helps to achieve better server consolidation resulting in better energy efficiency. Proposed model also relies on network load inside each data center which helps avoiding hotspots.

모듈러 데이터센터 냉각시스템의 취출온도 조건에 의한 연간에너지 소비량 예측

The modular data center is greatly influenced by operating conditions because its internal size is smaller than the existing hyperscale data center. The recent emergence of servers that can adapt well to indoor temperature rise requires a review of existing temperature conditions. In this study, the potential for temperature condition mitigation and annual energy performance were analyzed for recent installations of improved thermal performance servers in container-type distributed modular data centers. According to the analysis, the upper dry bulb temperature of the intake air for maintaining the recommended level, acceptable level A1, and acceptable level A2 was 14℃, 16℃, and 19℃ respectively. The annual energy savings predicted in the analysis was 3.1% and 9.2%, when the intake air temperature of the servers in the modular data center was mitigated to the A1 and A2 levels respectively.

Predicting Response Time of Concurrent Queries with Similarity Models

Predicting query response time is an essential task for managing database systems, especially in modern large distributed data centers that execute heterogeneous query workloads concurrently. The core of such a model is to quantify query interaction, which is neglected by the state of the art models. This paper proposes a novel model that estimates query response time based on the similarity of query mixes. We introduce a notion called query rating for constructing the feature vector of a query and developed a measure of the similarity between two query mixes. We propose a static similarity model to estimate the response time of a query by using that of the most similar query mixes containing the query. We also build a dynamic model based on the static model to predict the remaining execution time of a query on-the-fly whenever a new query mix forms. A scheduling method is proposed with the similarity models as the key enablement, which schedules a workload with minimum execution time. The experimental evaluation shows that our models perform approximately 12% and 35% of the actual response time on average for static and dynamic respectively, and the scheduler with our model outperforms, up to 2.9x, that with conventional models consistently.

Reliability of SSDs in Enterprise Storage Systems

This article presents the first large-scale field study of NAND-based SSDs in enterprise storage systems (in contrast to drives in distributed data center storage systems). The study is based on a very comprehensive set of field data, covering 1.6 million SSDs of a major storage vendor (NetApp). The drives comprise three different manufacturers, 18 different models, 12 different capacities, and all major flash technologies (SLC, cMLC, eMLC, 3D-TLC). The data allows us to study a large number of factors that were not studied in prior works, including the effect of firmware versions, the reliability of TLC NAND, and the correlations between drives within a RAID system. This article presents our analysis, along with a number of practical implications derived from it.

Качество обслуживания в программно-определяемых сетях для научных приложений: возможности и проблемы

Scientific applications require to process, analyze and transfer large volumes of data in the shortest possible time from distributed data sources. In order to improve their performance, it is necessary to provide them with specific QoS parameters. On the other hand, SDN is presented as a new paradigm of communications networks that facilitates the management of the communications infrastructure and consequently allows to dynamically incorporate QoS parameters to the applications running in this type of network. With both these paradigms in mind, we conducted this research to answer the following questions: Do scientific applications that are running in an SDN-Enabled distributed data centers improve their performance? Do they consider network QoS parameters for job scheduling? The methodology used was to consult articles in specialized databases containing the keywords SDN and for scientific applications: HPC and Big Data. Then, we analyzed the articles where these keywords intersect with some of the parameters related to QoS in communications networks. Also, we reviewed QoS proposals in SDN to identify the advances in this research area. The results of this paper are: i) QoS is an open issue to incorporate in scientific applications that are running in an SDN ii) we identified the challenges to join both these paradigms, and iii) we present a strategy to provide QoS to scientific applications that are being executed among SDN-Enabled distributed data centers.

DLFT: Data and Layout Aware Fault Tolerance Framework for Big Data Transfer Systems

Various scientific research organizations generate several petabytes of data per year through computational science simulations. These data are often shared by geographically distributed data centers for data analysis. One of the major challenges in distributed environments is failure; hardware, network, and software might fail at any instant. Thus, high-speed and fault tolerant data transfer frameworks are vital for transferring such large data efficiently between the data centers. In this study, we proposed a bloom filter-based data aware probabilistic fault tolerance (DAFT) mechanism that can handle such failures. We also proposed a data and layout aware mechanism for fault tolerance (DLFT) to effectively handle the false positive matches of DAFT. We evaluated the data transfer and recovery time overheads of the proposed fault tolerance mechanisms on the overall data transfer performance. The experimental results demonstrated that the DAFT and DLFT mechanisms exhibit a maximum of 10% and a minimum of 2% recovery time overhead at 80% and 20% fault points respectively. However, we observed minimum to negligible overhead with respect to the overall data transfer rate.

Minimizing Energy Cost for Green Data Center by Exploring Heterogeneous Energy Resource

With the deteriorating effects resulting from global warming in many areas, geographically distributed data centers contribute greatly to carbon emissions, because the major energy supply is fossil fuels. Considering this issue, many geographically distributed data centers are attempting to use clean energy as their energy supply, such as fuel cells and renewable energy sources. However, not all workloads can be powered by a single power sources, since different workloads exhibit different characteristics. In this paper, we propose a fine-grained heterogeneous power distribution model with an objective of minimizing the total energy costs and the sum of the energy gap generated by the geographically distributed data centers powered by multiple types of energy resources. In order to achieve these two goals, we design a two-stage online algorithm to leverage the power supply of each energy source. In each time slot, we also consider a chance-constraint problem and use the Bernstein approximation to solve the problem. Finally, simulation results based on real-world traces illustrate that the proposed algorithm can achieve satisfactory performance.

Task aware optimized energy cost and carbon emission-based virtual machine placement in sustainable data centers

Management of IT services is rapidly adapting to the cloud computing environment due to optimized service delivery models. Geo distributed cloud data centers act as a backbone for providing fundamental infrastructure for cloud services delivery. Conversely, their high growing energy consumption rate is the major problem to be addressed. Cloud providers are in a hunger to identify different solutions to tackle energy management and carbon emission. In this work, a multi-cloud environment is modeled as geographically distributed data centers with varying solar power generation corresponding to its location, electricity price, carbon emission, and carbon tax. The energy management of the workload allocation algorithm is strongly dependent on the nature of the application considered. The task deadline and brownout information is used to bring in variation in task types. The renewable energy-aware workload allocation algorithm adaptive to task nature is proposed with migration policy to explore its impact on carbon emission, total energy cost, brown and renewable power consumption.

Minimizing the operation cost of distributed green data centers with energy storage under carbon capping

The expensive cost and intermittent availability of renewable energy bring great challenges to its efficient utilization in green data centers. In this paper, we propose a new way to achieve an explicit trade-off between operational cost and carbon emission by dynamic storing off-site renewable energy in distributed data centers. We first formulate a constrained stochastic optimization problem for cost minimization of data centers. Then, by leveraging Lyapunov optimization theory, we design an online Carbon Capped Cost Minimization algorithm (CCCM) to achieve a near-optimal cost with rigorous mathematical proof. Specially, the decisions at each time slot are determined with an efficient iterative algorithm based on the Generalized Benders Decomposition (GBD) technique. Finally, extensive simulations are conducted to show the effectiveness of our algorithm. The results show that our algorithm can save about 6% total costs compared with the algorithm without offsite energy storage.

Quality of Service in Software Defined Networks for Scientific Applications: Opportunities and Challenges

Scientific applications requires to process, analyze and transfer large volumes of data in the shortest possible time from distributed data sources. In order to improve their performance, it is necessary to provide them with specific QoS parameters. On the other hand, SDN is presented as a new paradigm of communications networks that facilitates the management of the communications infrastructure and consequently allows to dynamically incorporate QoS parameters to the applications running in this type of network. With both these paradigms in mind, we conducted this research to answer the following questions: Do scientific applications that are running in an SDN-Enabled distributed data centers improve their performance? Do they consider network QoS parameters for job scheduling? The methodology used was to consult articles in specialized databases containing the keywords SDN and for scientific applications: HPC and Big Data. Then, we analyzed the articles where these keywords intersect with some of the parameters related to QoS in communications networks. Also, we reviewed QoS proposals in SDN to identify the advances in this research area. The results of this paper are: i) QoS is an open issue to incorporate in scientific applications that are running in an SDN ii) we identified the challenges to join both these paradigms, and iii) we present a strategy to provide QoS to scientific applications that are being executed among SDN-Enabled distributed data centers.

QoS and power network stability aware simultaneous optimization of data center revenue and expenses

Data center profits are directly impacted by electric power consumption, which in turn has critical environmental implications. Geo distributed data centers maximize profits by optimizing resource allocation and leveraging diverse power tariffs available in a deregulated market. However, a power network state apathetic profit maximization can destabilize the grid and hamper a data center's sustainability goals. In this paper, we formulate the data center profit maximization as a constrained optimization problem. We then propose a multi-level algorithm where the lower level scheme is based on evolutionary optimization. The algorithm simultaneously optimizes revenue and expenses while preserving QoS and power network stability. The proposed approach considers real-time demand-based energy price variations to orchestrate request routing and resource allocation. The approach is suitable for heterogeneous and homogeneous data center architectures. The proposed scheme is also thermal-aware and considers both computing and cooling consumption. Simulation results show that the proposed technique is effective; it achieves a higher profit over a broad range of price variations and data center utilization levels.

QoSComm: A Data Flow Allocation Strategy among SDN-Based Data Centers for IoT Big Data Analytics

When Internet of Things (IoT) big data analytics (BDA) require to transfer data streams among software defined network (SDN)-based distributed data centers, the data flow forwarding in the communication network is typically done by an SDN controller using a traditional shortest path algorithm or just considering bandwidth requirements by the applications. In BDA, this scheme could affect their performance resulting in a longer job completion time because additional metrics were not considered, such as end-to-end delay, jitter, and packet loss rate in the data transfer path. These metrics are quality of service (QoS) parameters in the communication network. This research proposes a solution called QoSComm, an SDN strategy to allocate QoS-based data flows for BDA running across distributed data centers to minimize their job completion time. QoSComm operates in two phases: (i) based on the current communication network conditions, it calculates the feasible paths for each data center using a multi-objective optimization method; (ii) it distributes the resultant paths among data centers configuring their openflow Switches (OFS) dynamically. Simulation results show that QoSComm can improve BDA job completion time by an average of 18%.

Забезпечення доступності інформації у центрі обробки даних військового призначення на основі балансу продуктивності системи

Informatization of the Armed Forces of Ukraine involves the creation of information infrastructure and implementation of elements of the Unified Automated Management System of the Armed Forces of Ukraine. Data Processing Centers (DPCs) should become the software and hardware basis and the main component of the information infrastructure of the Ministry of Defense of Ukraine. In this regard, the performance indicators of the data center, such as its capacity and response time are important. Among the many factors affected by these indicators, the availability of information is more important. Accessibility of information means the state in which users who have access rights to information can implement them without hindrance. Access rights include: the right to read, modify, store, copy, destroy information. The current practice of assessing the performance of a Distributed Data Center Information System (DDCI) has only indicative character. The most rational approach to obtaining calculations of DDCI is the use of prognostic models. Knowing the DDCI performance forecast, you can predict the behavior of the system in different operating conditions and ensuring the availability of information in Data Center. The aim of the article is to develop an approach to ensure the availability of information to the data center based on the assessment of the balance of system performance. Modeling a Distributed Information System by the key performance indicators confirms the truth of the proposed approach and allows you to clearly assess the relationship between key performance indicators, the balance of which is the basis for ensuring the availability of information.

Data placement in distributed data centers for improved SLA and network cost

Large-scale data-intensive applications provide services to users by routing service requests to geographically distributed data centers interconnected by Internet links. In order to achieve good reliability and data access latency performance, cloud service providers often simultaneously place multiple copies of the data in different data centers. The network communication required for updating the multiple data copies incurs an operational cost. At the same time, the penalty incurred by the Service Level Agreement (SLA) violation for data access from the data centers also imposes an operational cost on the service providers. In this paper, we tackle the problem of data placement in distributed data centers with the aim to minimize the operational cost incurred by delay SLA violation penalty and inter-data center network communication, assuming each data has K data replicas. We propose a K-level Cluster-based Data Placement algorithm (K-CDP) for the problem. The algorithm solves the linear programming relaxation and dual programming problems corresponding to the problem of minimizing SLA violation penalty cost caused by placing a replica of each data in a data center. Based on the obtained solutions, the algorithm clusters the data so that the data with similar placeable data centers form a data cluster. For the data in each cluster, the algorithm selects K data centers to minimize the operational cost. We prove that algorithm K-CDP is 2-approximation to the data placement problem. Our simulation results demonstrate that the proposed algorithm can effectively reduce the penalty cost incurred by delay SLA violation, the network communication cost, and the operational cost of data centers.

A request dispatching method for efficient use of renewable energy in fog computing environments

In the emerging era of Internet of Things (IoT), fog computing plays a critical role in serving delay-sensitive and location-aware applications. As a result, fog nodes are envisioned to be heavily deployed and form future distributed data centers. Powering fog nodes with green energy sources (such as solar and wind), not only helps in environmental and CO2 emission control but also paves the way towards a “sustainable IoT technology”. However, the downside of green energy is its variation and unpredictability, which needs to be engineered. In this paper, we use the Lyapunov optimization technique to derive algorithms for dynamic dispatching of the users’ requests among the nearby fog nodes and remote data centers. The proposed algorithms take into account the time constraints of the requests and maintain the system stability while efficiently utilize the available green energy sources. Exhaustive simulation results, based on solar radiation data supplied by the Australian Bureau of Meteorology, confirm the efficiency of the proposed algorithms. In particular, in terms of service time, the number of deadline misses and green energy utilization, the proposed algorithms outperform the state-of-the-art alternative up to 6%, 17% and 12%, respectively.