Workload Prediction Method Research Articles

DuCFF: A Dual-Channel Feature-Fusion Network for Workload Prediction in a Cloud Infrastructure

Cloud infrastructures are designed to provide highly scalable, pay-as-per-use services to meet the performance requirements of users. The workload prediction of the cloud plays a crucial role in proactive auto-scaling and the dynamic management of resources to move toward fine-grained load balancing and job scheduling due to its ability to estimate upcoming workloads. However, due to users’ diverse usage demands, the changing characteristics of workloads have become more and more complex, including not only short-term irregular fluctuation characteristics but also long-term dynamic variations. This prevents existing workload-prediction methods from fully capturing the above characteristics, leading to degradation of prediction accuracy. To deal with the above problems, this paper proposes a framework based on a dual-channel temporal convolutional network and transformer (referred to as DuCFF) to perform workload prediction. Firstly, DuCFF introduces data preprocessing technology to decouple different components implied by workload data and combine the original workload to form new model inputs. Then, in a parallel manner, DuCFF adopts the temporal convolution network (TCN) channel to capture local irregular fluctuations in workload time series and the transformer channel to capture long-term dynamic variations. Finally, the features extracted from the above two channels are further fused, and workload prediction is achieved. The performance of the proposed DuCFF’s was verified on various workload benchmark datasets (i.e., ClarkNet and Google) and compared to its nine competitors. Experimental results show that the proposed DuCFF can achieve average performance improvements of 65.2%, 70%, 64.37%, and 15%, respectively, in terms of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and R-squared (R2) compared to the baseline model CNN-LSTM.

A Novel Hybrid Model for Docker Container Workload Prediction

The emergence of containers dramatically simplifies and facilitates the development and deployment of applications. More and more enterprises deploy their applications on the container cloud platform. For cloud service providers, an effective container workload prediction method is a must to achieve efficient utilization of cloud resources. However, the existing methods are either rarely based on container load characteristics or cannot make accurate real-time predictions. In this paper, we propose a Docker container workload proactive prediction method using a hybrid model combining triple exponential smoothing and long short-term memory (LSTM), which not only can capture both short-term and long-term dependencies in container resource time series but also smooth the container resource utilization data. In order to improve the prediction accuracy of the hybrid model, those two single models are combined using the mean absolute percentage error (MAPE) method. Besides, we design a real-time Docker workload prediction system for the hybrid model. Our experiments show that the mean absolute percentage error of the hybrid model is decreased by an average of 3.24%, 12.18%, 13.42%, 43.45%, and 50.69% compared with the LSTM, the triple exponential smoothing, ES-ARIMA, Bayesian Ridge Regression and BiLSTM with an acceptable time and computational cost overhead.

Performance Analysis of Machine Learning Centered Workload Prediction Models for Cloud

The precise estimation of resource usage is a complex and challenging issue due to the high variability and dimensionality of heterogeneous service types and dynamic workloads. Over the last few years, the prediction of resource usage and traffic has received ample attention from the research community. Many machine learning-based workload forecasting models have been developed by exploiting their computational power and learning capabilities. This paper presents the first systematic survey cum performance analysis-based comparative study of diversified machine learning-driven cloud workload prediction models. The discussion initiates with the significance of predictive resource management followed by a schematic description, operational design, motivation, and challenges concerning these workload prediction models. Classification and taxonomy of different prediction approaches into five distinct categories are presented focusing on the theoretical concepts and mathematical functioning of the existing state-of-the-art workload prediction methods. The most prominent prediction approaches belonging to a distinct class of machine learning models are thoroughly surveyed and compared. All five classified machine learning-based workload prediction models are implemented on a common platform for systematic investigation and comparison using three distinct benchmark cloud workload traces via experimental analysis. The essential key performance indicators of state-of-the-art approaches are evaluated for comparison and the paper is concluded by discussing the trade-offs and notable remarks.

Tr-Predictior: An Ensemble Transfer Learning Model for Small-Sample Cloud Workload Prediction.

Accurate workload prediction plays a key role in intelligent scheduling decisions on cloud platforms. There are massive amounts of short-workload sequences in the cloud platform, and the small amount of data and the presence of outliers make accurate workload sequence prediction a challenge. For the above issues, this paper proposes an ensemble learning method based on sample weight transfer and long short-term memory (LSTM), termed as Tr-Predictor. Specifically, a selection method of similar sequences combining time warp edit distance (TWED) and transfer entropy (TE) is proposed to select a source domain dataset with higher similarity for the target workload sequence. Then, we upgrade the basic learner of the ensemble model two-stage TrAdaBoost.R2 to LSTM in the deep model and enhance the ability of the ensemble model to extract sequence features. To optimize the weight adjustment strategy, we adopt a two-stage weight adjustment strategy and select the best weight for the learner according to the sample error and model error. Finally, the above process determines the parameters of the target model and uses the target model to predict the short-task sequences. In the experimental validation, we arbitrarily select nine sets of short-workload data from the Google dataset and three sets of short-workload data from the Alibaba cluster to verify the prediction effectiveness of the proposed algorithm. The experimental results show that compared with the commonly used cloud workload prediction methods Tr-Predictor has higher prediction accuracy on the small-sample workload. The prediction indicators of the ablation experiments show the performance gain of each part in the proposed method.

Multi-workflow scheduling and resource provisioning in Mobile Edge Computing using opposition-based Marine-Predator Algorithm

Workflow Scheduling in Mobile Edge Computing (MEC) tries to allocate the best possible set of resources for the workflows, considering objectives such as deadline, cost, energy, Quality of Service (QoS), and so on. However, MEC may be under different workloads from the IoT and this may not have the required amount of resources to efficiently handle the workflows. To mitigate this problem, in this paper, we use proactive resource provisioning and workload prediction methods. For this purpose, we present a workload prediction method using a multilayer feed-forward Artificial Neural Network (ANN) model and apply its results for resource provisioning. Afterward, we present an opposition-based version of the Marine-Predator Algorithm (MPA) algorithm, denoted as OMPA. In this algorithm, we present a probabilistic opposition-based learning (OBL) method, which benefits from the OBL, Quasi OBL, and dynamic OBL methods. Afterward, the OMPA algorithm is used for training the multi-layer feed-forward ANN model and multi-workflow scheduling by taking into account factors such as the makespan and number of Virtual Machines (VMs). Extensive experiments conducted in the iFogSim simulator and on the NASA and Saskatchewan datasets indicate that the proposed scheme can achieve better results compared to other metaheuristic algorithms and scheduling schemes.

Research on the Construction of a Nursing Education Management Model Based on a Small Data-Driven Model and Its Application.

Based on the concept of responsible holistic nursing care, a whole-process dual-tutor nursing practice model is established and its application effects are explored. This paper firstly reviews the research progress of nursing workload prediction methods at home and abroad, in order to provide a reference for clinical nursing workers in China to choose a scientific, reasonable, and easy-to-use nursing workload prediction method. It is proposed to construct a nursing education management model based on small data to provide ideas and references for nursing education management to effectively predict the evolutionary trend of students' behaviour and improve the level of accurate services. The experimental group adopted a dual-tutor responsibility system for the whole-process nursing practice model, including a complete three-level supervision system: a dual-tutor teaching system, a PDCA responsibility system for continuous improvement, and a multichannel teacher-student interaction platform; the control group adopted the traditional nursing practice model.

CANFIS: A Chaos Adaptive Neural Fuzzy Inference System for Workload Prediction in the Cloud

As cloud-based applications become increasingly solicited by companies and individuals, the competition between cloud providers that offer cloud services keeps increasing. To win this competition, cloud providers must provide sufficient computing resources that will satisfy users’ demand for every request. Workload prediction has been investigated extensively to resolve this issue using various techniques. This paper presents a workload prediction method called CANFIS, which combines the Savitzky-Golay (SG) filter and Chaotic time series analysis with the Adaptive Neural Fuzzy Inference System (ANFIS) to make predictions of cloud workloads. The SG filter is used to clean data from noise and outliers, and chaotic analysis is used to investigate the chaotic nature of workload and to build the improved ANFIS model. The proposed method is evaluated using real workload traces from web applications (i.e., Wikipedia and NASA Kennedy traces) and cluster applications (i.e., CPU and Memory of Google cluster). Experimental results show that the proposed CANFIS model can improve prediction accuracy compared to existing techniques, including simple ANFIS, Auto-Regressive Integrated Moving Average (ARIMA), Support Vector Machine (SVM), Long Short Term Memory (LSTM), and Neural Networks based methods. A statistical analysis is also performed using the Friedman test, along with Finner post hoc analysis to verify the efficiency of the proposed prediction model.

Workload time series prediction in storage systems: a deep learning based approach

Storage workload prediction is a critical step for fine-grained load balancing and job scheduling in realtime and adaptive cluster systems. However, how to perform workload time series prediction based on a deep learning method has not yet been thoroughly studied. In this paper, we propose a storage workload prediction method called CrystalLP based on deep learning. CrystalLP includes workload collecting, data preprocessing, time series prediction, and data post-processing phase. The time series prediction phase is based on a long short-term memory network (LSTM). Furthermore, to improve the efficiency of LSTM, we study the sensitivity of the hyperparameters in LSTM. Extensive experimental results show that CrystalLP can obtain performance improvement compared with three classic time series prediction algorithms.

Towards Accurate Prediction for High-Dimensional and Highly-Variable Cloud Workloads with Deep Learning

Resource provisioning for cloud computing necessitates the adaptive and accurate prediction of cloud workloads. However, the existing methods cannot effectively predict the high-dimensional and highly-variable cloud workloads. This results in resource wasting and inability to satisfy service level agreements (SLAs). Since recurrent neural network (RNN) is naturally suitable for sequential data analysis, it has been recently used to tackle the problem of workload prediction. However, RNN often performs poorly on learning long-term memory dependencies, and thus cannot make the accurate prediction of workloads. To address these important challenges, we propose a deep Learning based Prediction Algorithm for cloud Workloads (L-PAW). First, a top-sparse auto-encoder (TSA) is designed to effectively extract the essential representations of workloads from the original high-dimensional workload data. Next, we integrate TSA and gated recurrent unit (GRU) block into RNN to achieve the adaptive and accurate prediction for highly-variable workloads. Using real-world workload traces from Google and Alibaba cloud data centers and the DUX-based cluster, extensive experiments are conducted to demonstrate the effectiveness and adaptability of the L-PAW for different types of workloads with various prediction lengths. Moreover, the performance results show that the L-PAW achieves superior prediction accuracy compared to the classic RNN-based and other workload prediction methods for high-dimensional and highly-variable real-world cloud workloads.

Short-Term Load Forecasting of Virtual Machines Based on Improved Neural Network

Cloud computing allows applications to share resources on physical machines in an independent manner so as to effectively increase the utilization of physical machines. To enable cloud service users to obtain high-quality services, it is significant for cloud service providers to reduce the time required to allocate and deploy virtual machines, as well as to provide time windows for the deployment of physical servers. To ensure demand is met, one effective implementation method is to predict the workload of virtual machines in the future, as this contributes to the efficient deployment of a data center's requests from virtual machines to dynamically changing physical servers. However, due to the complexity of virtual machine requests, virtual machine workload prediction remains a significant challenge at present; on the one hand, it is difficult for standard recurrent neural networks to capture long-term dependencies because of the disappearance of gradients, while on the other hand, the long short-term memory method cannot handle irregular intervals. Accordingly, a new virtual machine workload prediction method that is capable of managing irregular time intervals is proposed in this paper. The proposed method combines the amount and time intervals of historical virtual machine work requests in order to accurately predict the virtual machine's future workload using a fixed length of time as a unit. Experiments show that the proposed model can generate more accurate prediction results than the long short-term memory method.

StreamTune: dynamic resource scheduling approach for workload skew in video data center

Video surveillance applications need video data center to provide elastic virtual machine (VM) provisioning. However, the workloads of the VMs are hardly to be predicted for online video surveillance service. The unknown arrival workloads easily lead to workload skew among VMs. In this paper, we study how to balance the workload skew on online video surveillance system. First, we design the system framework for online surveillance service which consists of video capturing and analysis tasks. Second, we propose StreamTune, an online resource scheduling approach for workload balancing, to deal with irregular video analysis workload with the minimum number of VMs. We aim at timely balancing the workload skew on video analyzers without depending on any workload prediction method. Furthermore, we evaluate the performance of the proposed approach using a traffic surveillance application. The experimental results show that our approach is well adaptive to the variation of workload and achieves workload balance with less VMs.

Depth Assisted Adaptive Workload Balancing for Parallel View Synthesis

Depth image-based rendering has been adopted by MPEG as the recommended view synthesis technique for free viewpoint TV applications. In this paper, a workload balancing algorithm is proposed for parallel view synthesis on multicore platforms. First, view synthesis workload is defined as the function of the number of hole-pixels in the warped images. Then, a novel depth assisted prediction method is proposed to predict the number of hole-pixels in the current frame by exploiting the depth differences between the neighboring frames, which reflects the movement of objects in video content. Feeding the predicted workload to the proposed cost function, each input frame is partitioned adaptively to balance the synthesis workload among the cores. The proposed workload prediction method outperforms the existing approaches both in terms of frame average prediction error and standard deviation in prediction error. Applying the proposed workload balancing method, the parallel view synthesis system provides higher acceleration ratio and better synchronization performance among the cores compared with other parallel processing systems without sacrificing the subjective and objective quality. It is also robust to different platforms, which shows high potential in being applied to mobile oriented applications.

Adaptive VM Management with Two Phase Power Consumption Cost Models in Cloud Datacenter

As cloud computing models have evolved from clusters to large-scale data centers, reducing the energy consumption, which is a large part of the overall operating expense of data centers, has received much attention lately. From a cluster-level viewpoint, the most popular method for an energy efficient cloud is Dynamic Right Sizing (DRS), which turns off idle servers that do not have any virtual resources running. To maximize the energy efficiency with DRS, one of the primary adaptive resource management strategies is a Virtual Machine (VM) migration which consolidates VM instances into as few servers as possible. In this paper, we propose a Two Phase based Adaptive Resource Management (TP-ARM) scheme that migrates VM instances from under-utilized servers that are supposed to be turned off to sustainable ones based on their monitored resource utilizations in real time. In addition, we designed a Self-Adjusting Workload Prediction (SAWP) method to improve the forecasting accuracy of resource utilization even under irregular demand patterns. From the experimental results using real cloud servers, we show that our proposed schemes provide the superior performance of energy consumption, resource utilization and job completion time over existing resource allocation schemes.

Workload prediction using run‐length encoding for runtime processor power management

In processor power management research, workload prediction is a requisite for adjusting frequency without performance loss, and previous studies have proposed various prediction algorithms. Among them, methods based on the workload history table are lightweight and have high prediction accuracy for a variable workload. However, such prediction algorithms lose their prediction accuracy in the case of repeated workload. An improved workload prediction method using run-length encoding is proposed, which handles workload repetition. Evaluation results show that the proposed algorithm improves the prediction of repeated workload by up to 14% and also improves 4% of energy saving.

Cost-aware request routing in multi-geography cloud data centres using software-defined networking

Current geographically distributed cloud data centres (CDCs) require gigantic energy and bandwidth costs to provide multiple cloud applications to users around the world. Previous studies only focus on energy cost minimisation in distributed CDCs. However, a CDC provider needs to deliver gigantic data between users and distributed CDCs through internet service providers (ISPs). Geographical diversity of bandwidth and energy costs brings a highly challenging problem of how to minimise the total cost of a CDC provider. With the recently emerging software-defined networking, we study the total cost minimisation problem for a CDC provider by exploiting geographical diversity of energy and bandwidth costs. We formulate the total cost minimisation problem as a mixed integer non-linear programming (MINLP). Then, we develop heuristic algorithms to solve the problem and to provide a cost-aware request routing for joint optimisation of the selection of ISPs and the number of servers in distributed CDCs. Besides, to tackle the dynamic workload in distributed CDCs, this article proposes a regression-based workload prediction method to obtain future incoming workload. Finally, this work evaluates the cost-aware request routing by trace-driven simulation and compares it with the existing approaches to demonstrate its effectiveness.

A task-level adaptive MapReduce framework for real-time streaming data in healthcare applications

Healthcare scientific applications, such as body area network, require of deploying hundreds of interconnected sensors to monitor the health status of a host. One of the biggest challenges is the streaming data collected by all those sensors, which needs to be processed in real time. Follow-up data analysis would normally involve moving the collected big data to a cloud data center for status reporting and record tracking purpose. Therefore, an efficient cloud platform with very elastic scaling capacity is needed to support such kind of real time streaming data applications. The current cloud platform either lacks of such a module to process streaming data, or scales in regard to coarse-grained compute nodes.In this paper, we propose a task-level adaptive MapReduce framework. This framework extends the generic MapReduce architecture by designing each Map and Reduce task as a consistent running loop daemon. The beauty of this new framework is the scaling capability being designed at the Map and Task level, rather than being scaled from the compute-node level. This strategy is capable of not only scaling up and down in real time, but also leading to effective use of compute resources in cloud data center. As a first step towards implementing this framework in real cloud, we developed a simulator that captures workload strength, and provisions the amount of Map and Reduce tasks just in need and in real time.To further enhance the framework, we applied two streaming data workload prediction methods, smoothing and Kalman filter, to estimate the unknown workload characteristics. We see 63.1% performance improvement by using the Kalman filter method to predict the workload. We also use real streaming data workload trace to test the framework. Experimental results show that this framework schedules the Map and Reduce tasks very efficiently, as the streaming data changes its arrival rate.

Ensemble Learning for Large-Scale Workload Prediction

Increasing energy costs of large-scale server systems have led to a demand for innovative methods for optimizing resource utilization in these systems. Such methods aim to reduce server energy consumption, cooling requirements, carbon footprint, and so on, thereby leading to improved holistic sustainability of the overall server infrastructure. At the core of many of these methods lie reliable workload-prediction techniques that guide in identifying servers, time intervals, and other parameters that are needed for building sustainability solutions based on techniques like virtualization and server consolidation for server systems. Many workload prediction methods have been proposed in the recent paper, but unfortunately they do not deal adequately with the issues that arise specifically in large-scale server systems, viz., extensive nonstationarity of server workloads, and massive online streaming data. In this paper, we fill this gap by proposing two online ensemble learning methods for workload prediction, which address these issues in large-scale server systems. The proposed algorithms are motivated from the weighted majority and simulatable experts approaches, which we extend and adapt to the large-scale workload prediction problem. We demonstrate the effectiveness of our algorithms using real and synthetic data sets, and show that using the proposed algorithms, the workloads of 91% of servers in a real data center can be predicted with accuracy > 89%, whereas using baseline approaches, the workloads of only 13%–24% of the servers can be predicted with similar accuracy.

A User Preference and Service Time Mix-aware Resource Provisioning Strategy for Multi-tier Cloud Services

As the majority of cloud services deploy a multi-tier architecture, it poses a real challenge to resource provisioning. Because of the cross-tier dependencies of multi-tier cloud services, the traditional tailor-made resource provisioning methods for single-tier applications can’t be adopted or extended directly. Resource provisioning in the cloud is usually driven by performance predictions, so it's important to characterize workload fluctuations accurately in order to understand how to allocate resources. And user preference also contributes to workload variants. Therefore, we propose a dynamic resource provisioning strategy for multi-tier cloud services that employs (i) a user preference and service time mix-aware workload prediction method to be used as a foundation of resource provisioning, and (ii) a dynamic resource provisioning strategy based on the queuing theory to determine how many resources to be provisioned to each tier. Experimental results demonstrate that our workload prediction method is accurate, and our provisioning strategy is able to improve the accuracy of resource provisioning, reduce the allocated resources and SLAs violations.

Mental Workload Modeling in an Integrated Cognitive Architecture

Mental workload analysis is an important component in the test and evaluation of humanmachine systems. Existing empirical workload measures have limited applicability when humanin-the-loop tests are impractical, which produces the need for theory-based workload modeling and prediction methods. ACTR-QN is a theory-based integrated cognitive architecture combining the advantages of Adaptive Control of Thought-Rational (ACT-R) and Queueing Network (QN) architectures. The research reported in this paper proposes and examines a theory and method for modeling and visualizing mental workload in ACTR-QN. Validation with an empirical study of a semantic judgment task showed that an ACTR-QN model produced both performance and mental workload data similar to the human results. In addition, different components of the multidimensional mental workload can be visualized with ACTR-QN. Mental workload modeling in ACTR-QN provides a new tool for human factors evaluation of mental workload.

Predictive-Flow-Queue-Based Energy Optimization for Gigabit Ethernet Controllers

This paper presents energy-efficient packet interface architecture and a power management technique for gigabit Ethernet controllers, where low-latency and high-bandwidth are required to meet the pressing demands of very high frame-rate data. More specifically, a predictive-flow-queue (PFQ)-based packet interface architecture is presented, which adjusts the operating frequency of different functional blocks at a fine granularity so as to minimize the total system energy dissipation while attaining performance goals. A key feature of the proposed architecture is the implementation of a runtime workload prediction method for the network traffic along with a continuous frequency adjustment mechanism, which enables one to eliminate the latency and energy penalties associated with discrete power mode transitions. Furthermore, a stochastic modeling framework based on Markovian decision processes and queuing models is employed, which make it possible to adopt a precise mathematical programming formulation for the energy optimization under performance constraints. Experimental results with a designed 65-nm Gb Ethernet controller show that the proposed interface architecture and continuous frequency scaling result in system-wide energy savings while meeting performance specifications.