Search

The resources required to service cloud computing applications are dynamic and fluctuate over time in response to variations in the volume of incoming requests. Proactive autoscaling techniques attempt to predict future resource demand and assign resources before the request. Thus, establishing accurate prediction methods optimizes resource allocation and avoids service-level agreement violations (SLA). Most existing methods for predicting resource usage in cloud computing rely on univariate time series prediction models. These models consider only a single resource usage metric, make future predictions based on historical data for the target metric, and provide predictions for a single step ahead. This paper provides a hybrid method for multivariate time series workload prediction of host machines in cloud data centers that predicts the workload for the next few steps. First, a statistical analysis is used to construct the training set. Then, a convolutional neural network (CNN) is employed to extract the hidden spatial features between all correlated variables. Finally, the spatial features extracted by the CNN are fed into a GRU network optimized with the attention mechanism in order to extract the temporal correlation features. Two experiments employing Google cluster data were conducted to evaluate the proposed method. Experimental results reveal that our method improves prediction accuracy by 2% to 28% compared to baseline methods and previous research. [ABSTRACT FROM AUTHOR]

• Proactive auto-scaling method for container-based edge computing web applications. • Multi-step Federated Learning prediction model using BiGRU model in Edge Computing. • Reducing the data transmission between edge nodes and the cloud server. • Multi-step prediction in proactive auto-scaler approach in Kubernetes. Edge computing has emerged as an attractive alternative to traditional cloud computing by utilizing processing, network, and storage resources close to end devices, such as Internet of Things (IoT) sensors. Edge computing is still in its infancy, and resource provisioning and service scheduling remain research concerns. Kubernetes is a container orchestration tool in distributed environments. Proactive auto-scaling techniques in Kubernetes improve utilization by allocating resources based on future workload prediction. However, prediction models run on central cloud nodes, necessitating data transfer between edge and cloud nodes, which increases latency and response time. We present FedAvg-BiGRU, a proactive auto-scaling method in edge computing based on FedAvg and multi-step prediction by a Bidirectional Gated Recurrent Unit (BiGRU). FedAvg is a technique for training machine learning models in a Federated Learning (FL) model. FL reduces network traffic by exchanging only model updates rather than raw data, relieving learning models of the need to store data on a centralized cloud server. In addition, a technique for determining the number of Kubernetes pods based on the Cool Down Time (CDT) concept has been developed, preventing contradictory scaling actions. To our knowledge, our work is the first to employ FL for proactive auto-scaling in cloud and edge computing. The results demonstrate that the FedAvg-BiGRU method has a slightly higher prediction error than the load centralized processing mode, although this difference is not statistically significant. At the same time, it reduces the amount of data transmission between the edge nodes and the cloud server. [ABSTRACT FROM AUTHOR]

Massive amounts of real-time streaming network data are generated quickly because of the exponential growth of applications. Analyzing patterns in generated flow traffic streaming offers benefits in reducing traffic congestion, enhancing network management, and improving the quality of service management. Processing massive volumes of generated traffic poses more challenges when data traffic encryption is raised. Classifying encrypted network traffic in real-time with deep learning networks has received attention because of their excellent performance. The substantial volume of incoming packets, characterized by high speed and wide variety, puts real-time traffic classification within the domain of big data problems. Classifying traffic with high speed and accuracy is a significant challenge in the era of big data. The real-time nature of traffic intensifies deep learning networks, necessitating a considerable number of parameters, layers, and resources for optimal network training. Until now, various datasets have been employed to evaluate the effectiveness of previous methods for classifying encrypted traffic. The primary objective has been to enhance accuracy, precision, and F1-measure. Presently, encrypted traffic classification performance depends on pre-existing datasets. The learning and testing phases are done offline, and more research is needed to investigate the feasibility of these methods in real-world scenarios. This paper examines the possibility of a tradeoff between evaluating the model’s effectiveness, execution time, and utilization of processing resources when processing stream-based input data for traffic classification. We aim to explore the feasibility of establishing a tradeoff between these factors and determining optimal parameter settings. This paper used the ISCX VPN-Non VPN 2016 public dataset to evaluate the proposed method. All packets from the dataset were streamed continuously through Apache Kafka to the classification framework. Numerous experiments have been designed to demonstrate the efficacy of the proposed method. The experimental results show that the proposed method outperforms the baseline methods by 11% in the F1-measure when the number of workers is two and by 25% when the number of workers is equal to 32. [ABSTRACT FROM AUTHOR]

Abstract: The restricted energy of nodes is one of the most important challenges in wireless sensor networks. Since data transmissions among nodes consume most of the nodes' energy, thus, minimizing the unnecessary transmissions reduces the consumed energy. One of the sources of this problem is the redundancy of raw data that can be eliminated at the aggregation points. As a result, data aggregation can be considered as an effective strategy to tackle the mentioned issue and to optimize the communication energy consumption. In this paper, the sensor nodes are organized in a tree structure, and the data aggregation are done in intermediate nodes at the junction of tree branches. One of the main characteristics of tree protocols is reduction of energy consumption through optimizing the structure of a data aggregation tree. For this, this paper proposes to apply a swarm intelligent algorithm named river formation dynamics. The simulation results show that the proposed algorithm outperforms in comparison to the famous ant colony optimization algorithm in terms of network lifetime. Simulations show that the proposed algorithm makes nearly 4% and 50% improvement in lifetime of wireless sensor networks than ant colony optimization and shortest path routing, respectively. [ABSTRACT FROM AUTHOR]

Summary: Recently, many researches have been conducted to exploit the compressive sensing (CS) theory in wireless sensor networks (WSNs). One of the most important goals in CS is to prolong the lifetime of WSNs. But CS may suffer from some errors during the reconstruction phase. In addition, an adaptive version of CS named Bayesian compressive sensing has been studied to improve the reconstruction accuracy in WSNs. This paper investigates these adaptive methods and identifies their associated problems. Finally, a distributed and semi‐adaptive CS‐based data collection method is proposed. The proposed method tackles the aforementioned problems. Simulation results show that considering both lifetime and accuracy factors as a compound metric, the proposed method yields a 200% improvement compared with the Bayesian compressive sensing‐based method and outperforms other compared methods in the literature. [ABSTRACT FROM AUTHOR]

A wide range of applications in engineering and scientific computing are based on the sparse matrix computation. There exist a variety of data representations to keep the non-zero elements in sparse matrices, and each representation favors some matrices while not working well for some others. The existing studies tend to process all types of applications, e.g., the most popular application which is matrix-vector multiplication, with different sparse matrix structures using a fixed representation. While Graphics Processing Units (GPUs) have evolved into a very attractive platform for general purpose computations, most of the existing works on sparse matrix-vector multiplication (SpMV, for short) consider CPUs. In this work, we design and implement an adaptive GPU-based SpMV scheme that selects the best format for the input matrix having the configuration and characteristics of GPUs in mind. We study the effect of various parameters and different settings on the performance of SpMV applications when employing different data representations. We then employ an adaptive scheme to execute different sparse matrix applications using proper sparse matrix representation formats. Evaluation results show that our run-time adaptive scheme properly adapts to different applications by selecting an appropriate representation for each input sparse matrix. The preliminary results show that our adaptive scheme improves the performance of sparse matrix multiplications by 2.1 $$\times $$ for single-precision and 1.6 $$\times $$ for double-precision formats, on average. [ABSTRACT FROM AUTHOR]

Graphics Processing Units (GPUs) have a large and complex design space that needs to be explored in order to optimize the performance of future GPUs. Statistical techniques are useful tools to help computer architects to predict performance of complex processors. In this study, these methods are utilized to build a model which predicts the GPU performance efficiently. The design space of targeted Fermi GPU has more than 8 million points which cause exploring this huge design space a challenging process. In order to build an accurate model, we propose a two-tier algorithm in our algorithm which builds a multiple linear regression model from a small set of simulated data. In this algorithm the Plackett–Burman design is used to find the key parameters of the GPU, and further simulations are guided by a fractional factorial design for the most important parameters. Our algorithm is able to construct a GPU performance predictor which can predict the performance of any point in the design space with an average prediction error between 1% and 5% for different benchmark applications. In addition, in comparison to other methods which need a large number of sampling points, the accuracy in our method is achieved by only sampling between 0.0003% and 0.0015% of the full design space. [ABSTRACT FROM AUTHOR]

Understanding performance bottlenecks of applications in high-performance computing can lead to dramatic improvements in their performances. For example, a key problem in GPU programming is finding performance bottlenecks and solving them to reach the best possible performance. These bottlenecks in GPU architectures include a number of factors such as memory access latency, branch divergence, utilization, and the amount of existing parallelism. In addition, a simple profiling cannot demonstrate the relations between these bottlenecks. In this paper, we propose a statistical performance analyzer framework that not only helps us find bottlenecks, but also indicates the relations between them, which is not possible using a profiler. Recently, OpenCL has been proposed to be used in a variety of platforms, e.g., CPUs and GPUs, enabling a program written in one platform to be imported to other platforms with minimal effort. Therefore, we selected OpenCL to design our performance model for Nvidia GPUs. To construct the model, the values of GPU performance counters for the selected benchmarks are measured. Then, well-known statistical techniques such as regression and principal component analysis use these results to find the most significant parameters and to construct a performance model with up to 99 % accuracy. Finally, this method can be leveraged to characterize unknown applications based on their performance similarities with an existing database of benchmarks to predict their likely performance bottlenecks. [ABSTRACT FROM AUTHOR]

Mitigating energy consumption in Clouds reduces operational costs for providers. Power management policies which aim to reduce total energy consumed in data-centers pose challenges in both hardware technologies and resource management policies. We introduce an optimal utilization level of a host to execute a certain number of instructions to minimize energy consumption of the host. We also propose a virtual machine (VM) scheduling algorithm based on the unsurpassed utilization level to come up with optimal energy consumption while meeting a given QoS. In other words, our proposed algorithm regulates allocated computing resources of VMs on a host which results in reaching an optimal energy level in the host. The simulation results show that our proposed method not only reduces total energy consumption of a Cloud by 60 %, but also has a profound impact on turnaround times of real-time tasks by 94 %. It also increases the acceptance rate of arrival tasks by 96 %. Moreover, it takes a leading part in accepting lengthy tasks which have short deadlines. [ABSTRACT FROM AUTHOR]

Abstract: Communication overhead is the key obstacle to reaching hardware performance limits. The majority is associated with software overhead, a significant portion of which is attributed to message copying. To reduce this copying overhead, we have devised techniques that do not require to copy a received message in order for it to be bound to its final destination. Rather, a late-binding mechanism, which involves address translation and a dedicated cache, facilitates fast access to received messages by the consuming process/thread. We have introduced two policies namely Direct to Cache Transfer (DTCT) and lazy DTCT that determine whether a message after it is bound needs to be transferred into the data cache. We have studied the proposed methods in simulation and have shown their effectiveness in reducing access times to message payloads by the consuming process. [Copyright &y& Elsevier]

On-chip global nanophotonic interconnects are promising alternatives to traditional metallic interconnects in giga-scale integrated circuits. At the very first step, the design and optimization of nanophotonic interconnects require the modelling and characterization of optoelectronic devices with irregular 3D geometries. In this paper, a methodology for the SPICE-compatible modelling of such devices is proposed which does not rely on post-manufacturing data. This methodology couples the optics and electronics domains of such devices in order to improve the model accuracy. The proposed methodology has been applied to the modelling of a waveguide-coupled integrated Metal-Semiconductor-Metal PhotoDetector (MSM PD) with quadratic nanotapers. We have performed optics domain simulation in 3D field-solver MEEP and device simulations using Silvaco Atlas device-TCAD tool. The simulation results agree well with measurement data reported in literature. Agreement with less than 12% error at all bias voltages is observed. • This article presents a methodology that characterizes an optoelectronic device without relying on measurement data. • The presented methodology improves the modelling accuracy of metal-semiconductor-metal photodetectors. • The methodology does not rely on post-manufacturing data to make design space exploration possible prior to manufacturing. • The presented methodology couples the optics and electronics domains of functionality of the device. [ABSTRACT FROM AUTHOR]

During recent years, big data explosion and the increase in main memory capacity, on the one hand, and the need for faster data processing, on the other hand, have caused the development of various in-memory processing tools to manage and analyze data. Engaging the speed of the main memory and advantaging data locality, these tools can process a large amount of data with high performance. Apache Ignite, as a distributed in-memory platform, can process massive volumes of data in parallel. Currently, this platform is CPU-based and does not utilize the GPU's processing resources. To address this concern, we introduce Ignite-GPU that uses the GPU's massively parallel processing power. Ignite-GPU handles a number of challenges in integrating GPUs into Ignite and utilizes the GPU's available resources. We have also identified and eliminated time-consuming overheads and used various GPU-specific optimization techniques to improve overall performance. Eventually, we have evaluated Ignite-GPU with the Genetic Algorithm, as a representative of data and compute-intensive algorithms, and gained more than thousands of times speedup in comparison with its CPU version. [ABSTRACT FROM AUTHOR]

The utilization of cloud computing in Internet of Things (IoT) applications has become widespread. However, it presents challenges for latency-sensitive scenarios due to data transmission to the centralized cloud structure, which leads to increased network traffic and service delays. To address this, fog computing has emerged as an intermediary layer between the cloud and IoT, ensuring low-latency interactions. A pivotal challenge within the fog computing paradigm is the service placement problem, involving assigning services to appropriate nodes, which is recognized as NP-hard. Recognizing the intricate nature of service placement, this study introduces a multi-objective optimization approach tailored for dynamic service placement within container-based fog computing environments. Considering multiple objectives is imperative due to the complex interplay of performance metrics in fog computing scenarios. A two-tier resource management framework based on Kubernetes is proposed to manage these diverse yet interrelated objectives effectively. The framework harnesses the power of the multi-objective, non-dominated sorting genetic algorithm II (NSGA-II) to reconcile conflicting objectives and facilitate optimal service placement decisions. Incorporating multi-objective optimization enables a comprehensive evaluation of service placement solutions, ensuring a balanced trade-off between latency, cost-efficiency, and energy consumption. Empirical evaluations demonstrate that the proposed method improves cost, average latency time, and energy consumption by 8% to 36% compared to state-of-the-art methods. [ABSTRACT FROM AUTHOR]