Search

Large simulations with many independent sub-simulations are common in scientific computing. There are numerous challenges, however, associated with performing such simulations in shared computing environments. For example, sub-simulations may have wildly varying completion times or not complete at all, leading to unpredictable runtimes as well as unbalanced and inefficient use of human and computational resources. In this study, we use the actor model of concurrent computation to improve both the resource utilization and fault tolerance for large-scale scientific computing simulations. More specifically, we use actors in the SUMMA model to manage a large-scale hydrological simulation over the North American continent with over 500,000 independent sub-simulations. We find that the actors implementation outperforms a standard array job submission as well as the job submission tool GNU Parallel by better balancing the computational load across processors. The actors implementation also improves fault tolerance and can eliminate the user intervention required to detect and re-submit failed jobs. [ABSTRACT FROM AUTHOR]

This paper analyzes the tradeoffs between performance and resilience against cyber attacks of applications organized as workflows. The static nature of current workflows is a major benefit to attackers. To combat this advantage, a promising new approach inspired to Moving Target Defense (MTDs) was developed to increase a workflow's robustness to cyber attacks. This approach is based on dynamic reconfigurations of workflow tasks to reduce an attacker's probability of succeeding in completing the reconnaissance phase before launching an attack. Dynamic reconfigurations increase the resilience of a workflow against cyber attacks but increase its execution time due to the overhead of reconfigurations. As a part of this paper, we developed metrics that capture the impact of reconfigurations on a workflow's execution time and resilience against cyber attacks. The paper also presents recursive algorithms for computing the execution time and the reconnaissance function of a workflow. Our analysis relied on extensive trace-driven simulations of workflows from five different traces from the Workflow Trace Archive (WTA) and we used 6000 workflows from three different domains: scientific computing, engineering, and industrial. Our analysis of the results showed that there is a significant difference at the 95% confidence level due to reconfiguration on the resilience of workflows and demonstrated a consistent behavior across all five trace domains. [ABSTRACT FROM AUTHOR]

Large-scale scientific applications from various scientific domains (e.g., astronomy, physics, pharmaceuticals, chemistry, etc.) usually require substantial amounts of computing resources and storage space. International Grid computing resources can be a viable choice for supporting these challenging applications so that effectively locating suitable computing resources with minimal allocation overhead can be crucial. However, Grid resource availability is highly unstable and current Grid Information Service (GIS) cannot provide accurate state information. This can make it very difficult for users to schedule the jobs on the Grid system and to map tasks on appropriate available resources. In this paper, we present SCOUT system that can periodically profile Grid computing elements based on available number of CPU cores and average response time, and monitor the performance of each CE in the Virtual Organizations (VO). Micro-benchmark experimental results demonstrate that leveraging profiled data by SCOUT can improve the success rate of task executions and reduce the average response time. [ABSTRACT FROM AUTHOR]

This article presents an empirical evaluation of power consumption for scientific computing applications in multicore systems. Three types of applications are studied, in single and combined executions on Intel and AMD servers, for evaluating the overall power consumption of each application. The main results indicate that power consumption behavior has a strong dependency with the type of application. Additional performance analysis shows that the best load of the server regarding energy efficiency depends on the type of the applications, with efficiency decreasing in heavily loaded situations. These results allow formulating a model to characterize applications according to power consumption, efficiency, and resource sharing, which provide useful information for resource management and scheduling policies. Several scheduling strategies are evaluated using the proposed energy model over realistic scientific computing workloads. Results confirm that strategies that maximize host utilization provide the best energy efficiency and performance results. [ABSTRACT FROM AUTHOR]

Cloud Computing has become the preferred choice of performing scientific applications over the cloud since the computing has become as a utility. Cloud-based services have evolved exponentially and also the sophistication of Cloud infrastructure supporting these services has grown. Running traditional applications such as scientific data processing on Cloud, needs to consider the suitability of the could for the compliance of the Cloud critical metrics. This article aims to study the performance of public clouds to support the scientific computing, which were performed on Grid computing, High performance computing or cluster computing. This article tries to address the issue by designing and developing a framework to measure the critical metrics. The performance of the public clouds was assessed through probing, simulation and historic data. The results indicate that only a few provide heterogeneous cloud services. Performance of some of the clouds services can be improved by tweaking the critical metrics. [ABSTRACT FROM AUTHOR]

In this work, we have analyzed the input/output (I/O) activities of Cori, which is a high-performance computing system at the National Energy Research Scientific Computing Center at Lawrence Berkeley National Laboratory. Our analysis results indicate that most users do not adjust storage configurations but rather use the default settings. In addition, owing to the interference from many applications running simultaneously, the performance varies based on the system status. To configure file systems autonomously in complex environments, we developed DCA-IO, a dynamic distributed file system configuration adjustment algorithm that utilizes the system log information to adjust storage configurations automatically. Our scheme aims to improve the application performance and avoid interference from other applications without user intervention. Moreover, DCA-IO uses the existing system logs and does not require code modifications, an additional library, or user intervention. To demonstrate the effectiveness of DCA-IO, we performed experiments using I/O kernels of real applications in both an isolated small-sized Lustre environment and Cori. Our experimental results shows that our scheme can improve the performance of HPC applications by up to 263% with the default Lustre configuration. [ABSTRACT FROM AUTHOR]

SpMV (Sparse matrix-vector multiplication) is an essential component in scientific computing and has attracted the attention of researchers in related fields at home and abroad. With the continuous expansion of matrix data, the efficient parallel SpMV algorithm has become a research hotspot for research experts in related fields. The sparse matrix compression format as a critical point to improve computing performance can effectively save storage space and efficiently cooperate with the advantages of the processor system structure to give full play to performance. This paper proposes a new sparse matrix storage format CSR2 (Compressed Sparse Row 2). It is a new single format and suitable for processor platforms with SIMD (Single Instruction Multiple Data) vectorizations. The format operation of CSR2 is easy to implement with a low overhead of conversion. We compared the SpMV algorithm based on CSR2 with the most advanced single format CSR5 (Compressed Sparse Row 5) and Intel MKL (Intel Math Kernel Library) on the mainstream high-performance processor Intel Xeon E5-2670 v3 CPU. We choose 48 sets of matrices to be used as a benchmark suite. Experimental results show that CSR2 has a remarkable performance improvement compared with CSR5 and MKL. Compared to CSR5, CSR2 can achieve an average acceleration of 1.401 × (up to 1.861 ×). Compared to MKL, CSR2 can achieve an average acceleration of 1.261 × (up to 5.921 ×). In reality, for applications with multiple iterations, using our CSR2 can bring low-overhead format conversion and high-throughput computing performance. [ABSTRACT FROM AUTHOR]

Sparse triangular solve (SpTRSV) is an important kernel in many scientific computing applications. In traditional viewpoints, accelerating SpTRSV by parallelizing the solution process is a challenging task. Dependencies among the variables that exist in the solution process not only restrict the parallelism that can be achieved, but also introduce large synchronization overhead among the parallel tasks. Moreover, a time-consuming pre-processing phase is commonly required to identify calculations that can be parallelized. However, we have observed that a large number of dependencies among the variables can be eliminated if we only calculate partial values of the variables first and add them together to obtain the final values later. By using such a strategy, starting to solve a variable does not need to wait for all of its prerequisite variables having been solved. In consequence, parallelism of the SpTRSV can be increased significantly. In this paper, we transform above mentioned observations into a subtree-based parallel algorithm to accelerate SpTRSV. The proposed algorithm calculates partial values of the variable along with an implicit subtree traversal and utilizes hardware atomic operation to implement accumulation of the partial values. This not only introduces no pre-processing overhead, but also avoids any barrier synchronization among the parallel threads. We evaluate the proposed algorithm on 2135 matrices from SuiteSparse Matrix Collection based on a generic GPU platform. Experimental results demonstrate that our scheme outperforms the state-of-the-art GPU and CPU vendor libraries in 1949 and 1782 matrices, respectively. Compared with the latest synchronization-free method, our scheme outperforms in 1779 matrices. [ABSTRACT FROM AUTHOR]

With the popularization and development of cloud computing, lots of scientific computing applications are conducted in cloud environments. However, current application scenario of scientific computing is also becoming increasingly dynamic and complicated, such as unpredictable submission times of jobs, different priorities of jobs, deadlines and budget constraints of executing jobs. Thus, how to perform scientific computing efficiently in cloud has become an urgent problem. To address this problem, we design an elastic resource provisioning and task scheduling mechanism to perform scientific workflow jobs in cloud. The goal of this mechanism is to complete as many high-priority workflow jobs as possible under budget and deadline constraints. This mechanism consists of four steps: job preprocessing, job admission control, elastic resource provisioning and task scheduling. We perform the evaluation with four kinds of real scientific workflow jobs under different budget constraints. We also consider the uncertainties of task runtime estimations, provisioning delays, and failures in evaluation. The results show that in most cases our mechanism achieves a better performance than other mechanisms. In addition, the uncertainties of task runtime estimations, VM provisioning delays, and task failures do not have major impact on the mechanism's performance. [ABSTRACT FROM AUTHOR]

In many scientific and engineering areas there are emerging software services available over the Web. The reason for deploying such services in the Cloud is either to reduce the operational costs or to support the peaks in their usage profiles. The algorithms employed in such services are usually result of a long term research and technology development work, so it is beneficial to reuse those critical application parts when developing new Cloud applications. This paper investigates the possibilities to introduce a Model Driven Architecture (MDA) for the Cloud computing domain, which would support composition, customization, flexibility, maintenance and reusability of Cloud application components in the particular case of scientific and engineering applications. The underlying middleware technology of choice is the mOSAIC Platform as a Service (PaaS) solution. This choice is motivated by the fact that in mOSAIC a Cloud application consists of loosely coupled components, which are either generic and provide for key resource types needed by an application (computation, storage, communication) or custom made, e.g. based on existing legacy software. The MDA approach is illustrated through the design and operation of an application for analysis of structures under static loading. It is shown that a relatively simple design can be used to address two application bottlenecks: the varying number of users and the computational complexity of the given problem. The design reduces the necessary application development efforts and the key components can be reused for similar applications. [ABSTRACT FROM AUTHOR]

The science cloud paradigm has been actively developed and investigated, but still requires a suitable model for science cloud system in order to support increasing scientific computation needs with high performance. This paper presents an effective provisioning model of science cloud, particularly for large-scale high throughput computing applications. In this model, we utilize job traces where a statistical method is applied to pick the most influential features to improve application performance. With these features, a system determines where VM is deployed (allocation) and which instance type is proper (provisioning). An adaptive evaluation step which is subsequent to the job execution enables our model to adapt to dynamical computing environments. We show performance achievements by comparing the proposed model with other policies through experiments and expect noticeable improvements on performance as well as reduction of cost from resource consumption through our model. [ABSTRACT FROM AUTHOR]

I/O intensive applications have posed great challenges to computational scientists. A major problem of these applications is that users have to sacrifice performance requirements in order to satisfy storage capacity requirements in a conventional computing environment. Further performance improvement is impeded by the physical nature of these storage media even when state-of-the-art I/O optimizations are employed.In this paper, we present a distributed multi-storage resource architecture, which can satisfy both performance and capacity requirements by employing multiple storage resources. Compared to a traditional single storage resource architecture, our architecture provides a more flexible and reliable computing environment. This architecture can bring new opportunities for high performance computing as well as inherit state-of-the-art I/O optimization approaches that have already been developed. It provides application users with high-performance storage access even when they do not have the availability of a single large local storage archive at their disposal. We also develop an Application Programming Interface (API) that provides transparent management and access to various storage resources in our computing environment. Since I/O usually dominates the performance in I/O intensive applications, we establish an I/O performance prediction mechanism which consists of a performance database and a prediction algorithm to help users better evaluate and schedule their applications. A tool is also developed to help users automatically generate performance data stored in databases. The experiments show that our multi-storage resource architecture is a promising platform for high performance distributed computing. [ABSTRACT FROM AUTHOR]

In the world of high-performance computing, it is known that it can be beneficial to leave a small number of CPU cores unused, a practice termed undersubscription. However, undersubscription is rarely implemented in scientific applications of high-performance computing. We demonstrate the importance of calibrated undersubscription in Computational Fluid Dynamics simulations through the aggregated results of 1844 benchmarks. These benchmarks measured three hardware configurations and five different CFD models. On average, performance increased by 14% (weighted by node count). Performance improvements were most significant at large node counts, particularly when nearing a regime of negative scalability. We found that undersubscription could increase maximum performance by up to 50%; this advantage diminished as node count decreased but remained as much as 13% with a single node. In some cases, maximum performance was achieved with large numbers of free cores—nearly half of the cores in one case. Producing a regression from our dataset, we universally predict the optimal number of free cores as a function of cells per core. This regression achieves a 15% speed increase on average (again weighted by node count). [ABSTRACT FROM AUTHOR]

With the fast evolution of Internet of Things (IoT) applications, Wireless Sensor Networks (WSNs) have become a crucial part of modern infrastructure. The efficient provision of services and wiser use of resources are currently of great importance. WSNs consist of several sensor nodes that collaborate to monitor and send data to a central location known as a sink. The sink, also called a base station, serves as the endpoint for data transmission in each round. However, due to the limited computation, storage, and energy resources of sensor nodes, they often face challenges in changing clusters. Optimal selection of a node, aimed at minimizing network fragmentation and enhancing energy utilization, necessitates a sophisticated evaluation and computational procedure, demanding a substantial energy investment. Subsequently, the task at hand is the development of a system facilitating the connection of remote sensing sources to WSNs with minimal energy consumption. The primary goals of WSNs based on the IoT revolve around extending network longevity and enhancing energy efficiency. In the realm of IoT-based WSNs, where the efficiency of data collection and management is paramount, cluster-based methodologies have demonstrated their effectiveness. This investigation proposes the implementation of a most valuable player algorithm (MVPA) specifically tailored for IoT-based WSNs, taking into account diverse factors influencing node energy and network lifespan. The MVPA is a highly competitive optimization method that converges faster (with fewer function evaluations) and has a greater overall success rate. In this case, the optimum cluster head for an IoT-based WSN was chosen using an MVPA to maximize energy savings. Simulation results demonstrate that the recommended strategy, when compared to other current methods, increases the network lifetime by using the minimum amount of energy needed to function. [ABSTRACT FROM AUTHOR]

Message Passing Interface (MPI) is the de facto standard for parallel programming, and collective operations in MPI are widely utilized by numerous scientific applications. The efficiency of these collective operations greatly impacts the performance of parallel applications. With the increasing scale and heterogeneity of HPC systems, the network environment has become more complex. The network states vary widely and dynamically between node pairs, and this makes it more difficult to design efficient collective communication algorithms. In this paper, we propose a method to optimize collective operations by using real-time measured network states, specifically focusing on the binomial tree algorithm. Our approach employs a low-overhead method to measure the network states, and the binomial tree with small latency is constructed based on the measurement result. Additionally, we take into account the disparities between the two underlying MPI peer-to-peer communication protocols, eager and rendezvous, and design tailored binomial tree construction algorithms for each protocol. We have implemented hierarchical MPI_Bcast, MPI_Reduce, MPI_Gather and MPI_Scatter, utilizing our network states-aware binomial tree algorithm at the inter-node level. The benchmark results demonstrate that our algorithm effectively enhances performance in small and medium message communication when compared to the default binomial tree algorithm in Open MPI. Specifically, for MPI_Bcast, we observe an average performance improvement of over 15.5% when the message size is less than 64KB. Similarly, for MPI_Reduce, there is an average performance improvement of over 12.1% when the message is below 2KB. In addition, there is an average performance improvement of over 10% for MPI_Gather when the message ranging from 64B to 512B. For MPI_Scatter, our algorithm achieved performance improvement only for certain message sizes. [ABSTRACT FROM AUTHOR]

The pervasive adoption of Artificial Intelligence (AI) and Machine Learning (ML) applications has exponentially increased the demand for efficient resource allocation, workload scheduling, and parallel computing capabilities in cloud environments. This research addresses the critical need for enhancing both the scalability and security of AI/ML workloads in cloud computing settings. The study emphasizes the optimization of resource allocation strategies to accommodate the diverse requirements of AI/ML workloads. Efficient resource allocation ensures that computational resources are utilized judiciously, avoiding bottlenecks and latency issues that could hinder the performance of AI/ML applications. The research explores advanced parallel computing techniques to harness the full possible cloud infrastructure, enhancing the speed and efficiency of AI/ML computations. The integration of robust security measures is crucial to safeguard sensitive data and models processed in the cloud. The research delves into secure multi-party computation and encryption techniques like the Hybrid Heft Pso Ga algorithm, Heuristic Function for Adaptive Batch Stream Scheduling Module (ABSS) and allocation of resources parallel computing and Kuhn–Munkres algorithm tailored for AI/ML workloads, ensuring confidentiality and integrity throughout the computation lifecycle. To validate the proposed methodologies, the research employs extensive simulations and real-world experiments. The proposed ABSS_SSMM method achieves the highest accuracy and throughput values of 98% and 94%, respectively. The contributions of this research extend to the broader cloud computing and AI/ML communities. By providing scalable and secure solutions, the study aims to empower cloud service providers, enterprises, and researchers to leverage AI/ML technologies with confidence. The findings are anticipated to inform the design and implementation of next-generation cloud platforms that seamlessly support the evolving landscape of AI/ML applications, fostering innovation and driving the adoption of intelligent technologies in diverse domains. [ABSTRACT FROM AUTHOR]

Intrusion detection datasets are highly likely to contain numerous redundant, irrelevant, and noisy features that slow the performance of the machine learning techniques and classifiers that may be applied to them. The feature selection approach is used for reducing the number of features in intrusion detection datasets and eliminating those that are not important. One of the most powerful structured population approaches is the Cellular Automata approach, which is used to enhance the diversity and convergence of population-based optimization algorithms. In this work, the Cellular Automata approach, Mixed opposition-based learning, and the K-Nearest Neighbor classifier are incorporated with the Prairie dog optimization algorithm (PDO) in a new intrusion detection framework called Binary Opposition Cellular Prairie dog optimization algorithm (BOC-PDO). The proposed framework contains four key features. First, the Cellular Automata model is utilized to enhance the population of feasible solutions in the PDO. Second, four S-shaped and four V-shaped Binary Transfer Functions are used to convert the continuous solutions in BOC-PDO to binary ones. Third, the Mixed opposition-based learning approach is used at the end of the optimization loop of BOC-PDO to improve capacity for exploration. Fourth, the K-Nearest Neighbor classifier is used as the main learning model in BOC-PDO. Eleven famous intrusion detection datasets were employed in the evaluation of the effectiveness of BOC-PDO compared to eight popular binary optimization algorithms and four machine learning approaches. According to the overall simulation results, BOC-PDO scored the highest accuracy, best objective value, and fewest selected features for each of the eleven intrusion detection datasets. Besides, the reliability and consistency of the simulation results of BOC-PDO compared to the other tested algorithms were established using Friedman and Wilcoxon statistical tests. [ABSTRACT FROM AUTHOR]

Features within the dataset carry a significant role; however, resource utilization, prediction-time, and model weight are increased by utilizing high-dimensional data in intrusion-detection paradigm. This paper aims to design a novel lightweight intrusion detection system in two phases utilizing a swarm intelligence-based technique. In 1st-phase, essential features are selected using particle swarm optimization algorithm by considering imbalanced dataset. Ant colony optimization algorithm is utilized in 2nd-phase for extracting information-rich and uncorrelated features. Additionally, genetic algorithm is employed for fine-tuning each detection model. Proposed model's performance is evaluated on different base and ensemble classifiers, and it is observed that xgboost achieves best accuracy with 90.38%, 92.63%, and 97.87% on NSL-KDD, UNSW-NB15, and CSE-CIC-IDS2018 datasets, respectively. The proposed model also outperforms other traditional dimensionality reduction and state-of-the-art approaches with statistical validation. This paper also analyses objective function of each metaheuristic algorithm used in this paper, applying convergence graphs, box, and swarm plots. [ABSTRACT FROM AUTHOR]

The Sparrow Search Algorithm (SSA) is recognized for its rapid convergence and precision in engineering optimization, yet it faces the challenge of premature convergence on complex problems. To address this, a multi-strategy improved sparrow search algorithm (MISSA) is proposed to enhance the optimization performance and applicability in this study. For the first time in the algorithm, the first definition of ellipses is integrated into SSA to balance its exploration and exploitation capabilities. A group co-evolutionary mechanism is introduced to promote population diversity and suppress premature convergence. Unlike most existing work, ablation experiments are utilized to evaluate the effective impact of these enhancement strategies on SSA. Statistical results based on the Wilcoxon signed-rank test and Friedman test show that the dynamic regulator based on the first definition of ellipses has the greatest impact on improving the performance of SSA. Numerical experiments based on the CEC2017 benchmark problems are used as an optimization case to compare MISSA with the classical metaheuristic algorithm and other state-of-the-art variants of SSA. The results demonstrate the outstanding performance and immense potential of MISSA in problem-solving. The applicability of the proposed algorithm is validated through six actual engineering optimization problems, showcasing strong competitiveness in global optimization. [ABSTRACT FROM AUTHOR]

Conventional signcryption schemes that depend on discrete logarithm problem, big integer prime factorization, and bilinear pairing are at risk in the context of quantum computers. We suggest a signcryption scheme based on a lattice to overcome the problems posed due to potential quantum computers in healthcare management which is named as lattice-based ring signcryption scheme for smart healthcare management (LRS-SHM). Unlike existing schemes, our design uses regenerated keys for every signature generated by user itself (instead of a key generation center), thus protecting the privacy of the user. Furthermore, we provide an anonymous health record management scheme that combines a (t, n) threshold method with our lattice-based signcryption system in an effort to overcome the existing anonymity limits in health record administration, especially in light of quantum computing concerns. We show the anonymity of our lattice-based signcryption method against confidentiality and its existential non-forgeability against insider corruption in the context of health record administration, proving security under the random oracle model. We compare our proposed scheme with existing schemes and found that our scheme is more efficient than others. By the use of (t, n) threshold scheme, we protect our scheme from single point of failure because we distribute the secret share to n number of users and to reconstruct the secret the scheme requires at least t users. Empowering privacy, our lattice-based signcryption prevents quantum threats, introduces seamless key control, and enhances health record anonymity, setting a new standard for secure and efficient cryptographic solutions. [ABSTRACT FROM AUTHOR]

A distributed archetype, the concept of fog computing relocates the storage, computation, and services closer to the network's edge, where the data is generated. Despite these advantages, the users expect proper load management in the fog environment. This has expanded the Internet of Things (IoT) field, increasing user requests for the fog computing layer. Given the growth, Virtual Machines (VMs) in the fog layer become overburdened due to user demands. In the fog layer, it is essential to evenly and fairly distribute the workload among the segment's current VMs. Numerous load-management strategies for fog environments have been implemented up to this point. This study aims to create a hybridized and optimized approach for load management (HOGWO), in which the population set is generated using the Invasive Weed Optimisation (IWO) algorithm. The rest of the functional part is done with the help of the Grey Wolf Optimization (GWO) algorithm. This process ensures cost optimization, increased performance, scalability, and adaptability to any domain, such as healthcare, vehicular traffic management, etc. Also, the efficiency of the enhanced approach is analyzed in various scenarios to provide a more optimal solution set. The proposed approach is well illustrated and outperforms the existing algorithms, such as Particle Swarm Optimization (PSO), Genetic Algorithm (GA), etc., in terms of cost and load management. It was found that more than 97% jobs were completed on time, according to the testing data, and the hybrid technique outperformed all other approaches in terms of fluctuation of load and makespan. [ABSTRACT FROM AUTHOR]

The Network-on-Chip (NoC) is a communication infrastructure designed to integrate various components of a System-on-Chip (SoC) and connect multi-core processors. In on-chip networks, routing is a process that determines how a data packet should move from the source node to the destination node. Essentially, a routing algorithm configures the internal connections of a switch to establish communication towards the destination node by examining the source, destination and the current node addresses. After routing, multiple input channels may simultaneously request access to an output channel based on the routing results. In such cases, an arbitration strategy needs to prioritize access to the output channel for the requesting input channels. This research presents an arbitration strategy called HTPA (Hybrid Traffic Pattern-Aware arbitration) that utilizes local and global information. HTPA also uses a hybrid arbitration scheme consisting of a simple arbitration and a complex switch arbitration. Additionally, it uses age-related techniques to increase the fairness of arbitration and prevent starvation. The proposed method considers certain priorities for accessing output channels based on the distance to the destination, the number of available paths from the current node to the destination, and the traffic intensity parameter. Evaluations were conducted under uniform, transient, hot spot, and MPEG4 traffic patterns and the proposed method demonstrated a reduction in average packet delay compared to RoundRobin, CAIS, and DTIS methods by 79.46%, 71.35%, and 19.91%, respectively, across various traffic patterns. [ABSTRACT FROM AUTHOR]

Scientific workflows are essential for many applications, enabling the configuration and execution of complex tasks across distributed resources. In this paper, we contribute an Ethereum blockchain-based scientific workflow execution manager, which distributes workflows to run on cluster computing providers that utilize the Slurm workload manager to execute them. We extended our blockchain-based autonomous resource broker called eBlocBroker, which is a DAO-based decentralized coordinator, by providing distributed workflow execution via blockchain. Through various tests, we demonstrate how our eBlockBroker autonomous organization, which is programmed as a smart contract, can manage scientific workflow submission, scheduling, and execution on cluster computing providers. The utilization of blockchain for distributed workflow execution is a new concept. We are motivated because our system has been developed with e-Science in mind where scientific workflows are widely utilized. [ABSTRACT FROM AUTHOR]

Load balancing plays a critical role in ensuring system stability and optimal performance, and as such, it has been a subject of extensive research across diverse computing domains, particularly in heterogeneous systems. Such systems integrate various computing devices with distinct architectures and computational power, each designed to execute a particular type of workload having diverse and immediate requirements. This research introduces a novel load balancing approach that utilizes machine learning to monitor the load distribution of heterogeneous systems in real-time. It estimates the load of the system's devices according to our proposed definition of the load. This definition provided us with an accurate measurement of the actual load, enabling our approach to detect imbalances and locate underloaded devices. Subsequently, our approach assigns applications to these underloaded devices to achieve load balance. We conducted a comparative analysis between our proposed approach and various established load balancing methods from diverse computing domains. This evaluation included three critical criteria: load balancing, workload execution time, and energy consumption. Additionally, we introduced a novel metric called "IMBALANCE" that quantifies the difference in load distribution among the system devices. We implemented each load balancing approach in a real-time heterogeneous system consisting of a master host and three worker servers each equipped with a Central Processing Unit and Graphical Processing Unit. To ensure a comprehensive validation of our approach, we conducted experiments in three distinct scenarios, involving different types of workloads having varying levels of computing intensity. The experimental results consistently demonstrate significant performance improvements of our approach across all scenarios. Our approach excels in minimizing system imbalance, which, in turn, enhances load balancing. This leads to reduced workload execution times and minimized device idle periods, ultimately improving energy efficiency. [ABSTRACT FROM AUTHOR]

Cloud computing has gained significant popularity, as a promising service platform in recent years. It facilitates the users to offload their resource-intensive computations to cloud server (CS) for efficient processing. Meanwhile, users prioritize the security and privacy of their sensitive data by encrypting it before outsourcing it to the CS. However, performing computations like multivariate polynomial function over encrypted data and verifying their correctness poses a major challenge. Homomorphic Encryption (HE) has been proposed as a solution to enable computations over encrypted data while maintaining data confidentiality. This work proposes a secure and efficient scheme for verifiable computation on outsourced data, named SEVCOD. SEVCOD combines the power of multivariate polynomial factorization and homomorphic encryption to enable public verification of CS's computation results preserving confidentiality of both data and result. The effectiveness of our scheme is validated through security proofs and performance analysis. The result analysis demonstrates that, SEVCOD significantly reduces computational burden for verifiers and enhances the overall efficiency of the outsourced computation process. [ABSTRACT FROM AUTHOR]

This research proposes a novel Hybrid Success History Intelligent Optimizer with Gaussian Transformation (SHIOGT) for solving different complexity level optimization problems and for Convolutional Neural Network (CNNs) hyperparameter tuning. SHIOGT algorithm is designed to balance exploration and exploitation phases through the addition of Gaussian Transformation to the original Success History Intelligent Optimizer. The inclusion of Gaussian Transformation enhances solution diversity enables SHIO to avoid local optima. SHIOGT also demonstrates robustness and adaptability by dynamically adjusting its search strategy based on problem characteristics. Furthermore, the combination of Gaussian and SHIO facilitates faster convergence, accelerating the discovery of optimal or near-optimal solutions. Moreover, the hybridization of these two techniques brings a synergistic effect, enabling SHIOGT to overcome individual limitations and achieve superior performance in hyperparameter optimization tasks. SHIOGT was thoroughly assessed against an array of benchmark functions of varying complexities, demonstrating its ability to efficiently locate optimal or near-optimal solutions across different problem categories. Its robustness in tackling multimodal and deceptive landscapes and high-dimensional search spaces was particularly notable. SHIOGT has been benchmarked over 43 challenging optimization problems and have been compared with state-of-the art algorithm. Further, SHIOGT algorithm is applied to the domain of deep learning, with a case study focusing on hyperparameter tuning of CNNs. With the intelligent exploration–exploitation balance of SHIOGT, we hypothesized it could effectively optimize the CNN's hyperparameters. We evaluated the performance of SHIOGT across a variety of datasets, including MNIST, Fashion-MNIST, CIFAR-10, and CIFAR-100, with the aim of optimizing CNN model hyperparameters. The results show an impressive accuracy rate of 98% on the MNIST dataset. Similarly, the algorithm achieved a 92% accuracy rate on Fashion-MNIST, 76% on CIFAR-10, and 70% on CIFAR-100, underscoring its effectiveness across diverse datasets. [ABSTRACT FROM AUTHOR]

Wavelet neural networks (WNN) have been applied in many fields to solve regression as well as classification problems. After the advent of big data, as data gets generated at a brisk pace, it is imperative to analyze it as soon as it is generated owing to the fact that the nature of the data may change dramatically in short time intervals. This is necessitated by the fact that big data is all pervasive and throws computational challenges for data scientists. Therefore, in this paper, we built an efficient Scalable, Parallelized Wavelet Neural Network (SPWNN) which employs the parallel stochastic gradient algorithm (SGD) algorithm. SPWNN is designed and developed under both static and streaming environments in the horizontal parallelization framework. SPWNN is implemented by using Morlet and Gaussian functions as activation functions. This study is conducted on big datasets like gas sensor data which has more than 4 million samples and medical research data which has more than 10,000 features, which are high dimensional in nature. The experimental analysis indicates that in the static environment, SPWNN with Morlet activation function outperformed SPWNN with Gaussian on the classification datasets. However, in the case of regression, there is no clear trend was observed. In contrast, in the streaming environment i.e., Gaussian outperformed Morlet on the classification and Morlet outperformed Gaussian on the regression datasets. Overall, the proposed SPWNN architecture achieved a speedup of 1.22 - 1.78. [ABSTRACT FROM AUTHOR]

Reducing power consumption with tolerable performance degradation is a fundamental challenge in today's containerized High Performance Computing (HPC). Most Dynamic Power Management (DPM) approaches proposed for the HPC environment are based on profile-guided power-performance prediction techniques. However, the complexity of DPM approaches in a multi-tenant containerized HPC environment (cHPCe) increases significantly due to the varying demands of users and the contention of shared resources. Moreover, there is limited research into software-level monitoring of power consumption in the popular Docker container environment since it is not designed keeping HPC in mind. The proposed research in this paper aims to present a real-time hybrid power-performance prediction approach using Long Short-Term Memory (LSTM) machine learning model. Unlike state-of-the-art techniques, LSTM with the rolling update mechanism accurately predicts the relationships between the tail latencies in time-series power performance data. It updates the training sample sequences according to the currently predicted power consumption and duration to predict sporadic power surges. It also proposes a power-cap determination framework with resource contention awareness to fine-tune power consumption in real time at the thread level. The proposed containerized environment is designed from scratch keeping HPC requirements in mind. Hence, we name the proposed approach as power-aware cHPCe (pcHPCe) and is evaluated and compared with native BareMetal execution using the NAS Parallel Benchmark (NPB) and HPC Challenge benchmark (HPCC) applications. Our experimental results show that the power-performance prediction model achieves an accuracy of 91.39% on average in real-time with an overhead of 1.6% of the total computing power per node. Our resource contention-aware power-cap selection framework attains significant power saving up to 13.1%. [ABSTRACT FROM AUTHOR]

Abstract  Computational simulations and thus scientific computing is the third pillar alongside theory and experiment in todays science. The term e-science evolved as a new research field that focuses on collaboration in key areas of science using next generation computing infrastructures (i.e. co-called e-science infrastructures) to extend the potential of scientific computing. During the past years, significant international and broader interdisciplinary research is increasingly carried out by global collaborations that often share a single e-science infrastructure. More recently, increasing complexity of e-science applications that embrace multiple physical models (i.e. multi-physics) and consider a larger range of scales (i.e. multi-scale) is creating a steadily growing demand for world-wide interoperable infrastructures that allow for new innovative types of e-science by jointly using different kinds of e-science infrastructures. But interoperable infrastructures are still not seamlessly provided today and we argue that this is due to the absence of a realistically implementable infrastructure reference model. Therefore, the fundamental goal of this paper is to provide insights into our proposed infrastructure reference model that represents a trimmed down version of ogsa in terms of functionality and complexity, while on the other hand being more specific and thus easier to implement. The proposed reference model is underpinned with experiences gained from e-science applications that achieve research advances by using interoperable e-science infrastructures. [ABSTRACT FROM AUTHOR]

Mutation testing is a valuable technique for measuring the quality of test suites in terms of detecting faults. However, one of its main drawbacks is its high computational cost. For this purpose, several approaches have been recently proposed to speed-up the mutation testing process by exploiting computational resources in distributed systems. However, bottlenecks have been detected when those techniques are applied in large-scale systems. This work improves the performance of mutation testing using large-scale systems by proposing a new load distribution algorithm, and parallelising different steps of the process. To demonstrate the benefits of our approach, we report on a thorough empirical evaluation, which analyses and compares our proposal with existing solutions executed in large-scale systems. The results show that our proposal outperforms the state-of-the-art distribution algorithms up to 35% in three different scenarios, reaching a reduction of the execution time of—at best—up to 99.66%. [ABSTRACT FROM AUTHOR]

Cloud computing platforms have been continuously evolving. Features such as the Elastic Fabric Adapter (EFA) in the Amazon Web Services (AWS) platform have brought yet another revolution in the High Performance Computing (HPC) world, further accelerating the convergence of HPC and cloud computing. Other public clouds also support similar features further fueling this change. In this paper, we show how and why the performance of a large-scale computational fluid dynamics (CFD) HPC application on AWS competes very closely with the one on Beskow—a Cray XC40 supercomputer at the PDC Center for High-Performance Computing - in terms of cost-efficiency with strong scaling up to 2304 processes. We perform an extensive set of micro and macro benchmarks in both environments and conduct a comparative analysis. Until as recently as 2020 these benchmarks have notoriously yielded unsatisfactory results for the cloud platforms compared with on-premise infrastructures. Our aim is to access the HPC capabilities of the cloud, and in general to demonstrate how researchers can scale and evaluate the performance of their application in the cloud. [ABSTRACT FROM AUTHOR]

With the increasing number of ransomware attacks on critical infrastructures, there is an urgent need to develop effective systems that can detect ransomware early. In order to achieve this objective, many detection solutions rely on machine learning to analyze the features of ransomware samples. However, these solutions often need to execute ransomware to extract sufficient features, which can increase the risk of virus infection. This paper proposes a novel static analysis framework based on the portable executable header, which utilizes deep reinforcement learning for early detection of ransomware. The framework leverages the portable executable header (PE header) from executable files as the key feature to identify ransomware. Moreover, the framework can learn the features of ransomware by using deep reinforcement learning. This is achieved through the interaction between the agent and the environment, and then the samples are segmented into ransomware and benign categories by taking actions. The proposed framework achieves rapid detection speeds without running ransomware samples by employing a lightweight network and the portable executable header from the raw byte files. To the best of our knowledge, we are the first to exploit deep reinforcement learning on the PE header for ransomware early detection. Our experiments on two real-world datasets reveal that the proposed framework can rapidly detect unseen ransomware. [ABSTRACT FROM AUTHOR]

With the rapid development of cloud computing, numerous large-scale workflow are executed in the cloud environment. Therefore, the workflow scheduling in cloud environment has become an emerging topic. This paper focuses on a reliability constrained multi-objective workflow scheduling problem (RCMOWSP) with the objectives of minimum execution cost and time. To solve the RCMOWSP, this paper proposes a knowledge-based multi-objective estimation of distribution algorithm (KMOEDA) with several problem-specific operators. First, an idle time-based decoding scheme is applied to sort the permutation of tasks greedily. In the global search strategy, a probability model is constructed to improve the diversity of population. Based on the problem-specific knowledge, a reliability-aware local search strategy is designed to performs local search around the solutions that violate reliability constraint. An elite enhancement strategy with a task perturbation operator and a resource perturbation operator is introduced to further improve the elite non-dominated solutions in the external archive. A comprehensive experiment is conducted to verify the performance of KMOEDA. The comparative results show that the KMOEDA significantly outperforms several relative multi-objective workflow scheduling approaches in solving the RCMOWSP. [ABSTRACT FROM AUTHOR]

In recent years, serverless computing has emerged as a new cloud service model that provides an elegant computing framework for application development. It allows developers to focus solely on application logic without having to worry about the underlying infrastructure or execution environment. One of the most attractive features of serverless computing is scale-to-zero, where function instances are immediately released as soon as they complete their execution. This makes the computing model more cost-effective. However, when a new function request arrives, a container must be prepared from scratch, including initialization, configuration, and loading of required libraries and packages. This requires additional startup time, also known as cold-start delay, which can degrade the performance of the computing framework. This paper proposes an integrated model, called adaptive container provisioning model (ACPM), to reduce cold-start latency through runtime provisioning of containers. ACPM promises to reduce the frequency of cold-starts and delay due to cold-start, improving the overall performance of serverless execution. It consists of two phases. In the first phase, an efficient deep learning model, such as long short term memory (LSTM), is used to predict the required number of pre-warmed containers in advance, which can be served as soon as the function request arrives. This phase is integrated with a container placement module for faster delivery of a container to minimize the frequency of cold-start delay. The second phase is based on a sandboxing mechanism where groups of similar functions can be placed in a single container instead of creating a separate container for each incoming request. ACPM drastically reduces cold-start delay and improves overall performance. The effectiveness of ACPM has been extensively compared with standard approaches such as container-sharing, container-reuse, pre-warming, restore-based, etc. From the experiment, it can be observed that ACPM outperforms other existing models in reducing cold-start latency. [ABSTRACT FROM AUTHOR]

Cloud computing has experienced extraordinary development across a wide range of industries by giving customers the flexibility to employ computing resources as needed. The task scheduling problem is one of several major challenges in cloud computing, and it should be scheduled effectively to minimize makespan and maximize resource utilization. Therefore, this paper put forward an improved scheduling efficiency algorithm called Enhanced Marine Predator Algorithm (EMPA). Firstly, task scheduling model with makespan and resource utilization is constructed. Secondly, each individual represents a result of task scheduling, and the purpose of algorithms is to find the optimal scheduling result, therefore the operator of WOA, nonlinear inertia weight coefficient and golden sine strategy are introduced into Marine Predator Algorithm. In the simulation experiment, EMPA is compared with Grey Wolf Optimizer (GWO), Sine Cosine Algorithm (SCA), Particle Swarm Optimization (PSO), and Whale Optimization Algorithm (WOA) under different number of tasks in synthetic datasets and GoCJ datasets.The experimental results show that the EMPA algorithm has more advantages in terms of makespan, degree of imbalance, and resource utilization. [ABSTRACT FROM AUTHOR]

One of the most challenging problems in cloud datacenters is the degradation of performance and energy efficiency due to the overutilization of hosts and their exposition to excessive workload. Virtual machine (VM) consolidation and migration from one host to another are strategies that have been proven to successfully bring about performance improvements and energy efficiency. These schemes help in energy optimization by moving VMs experiencing difficulty functioning on an overloaded host to another host. Similarly, by migrating VMs from an underloaded host and consolidating them, unnecessary resources have a chance to be shut down. This makes clear why the accurate detection of overloaded and underloaded hosts is of fundamental importance when energy consumption, quality of services, and service level agreements are targeted. In this paper, an energy-aware QoS-based consolidation algorithm is proposed to dynamically manage VMs in cloud datacenters. The proposed algorithm applies reinforcement learning and artificial neural networks. The first method is used to select a suitable VM for migration, while the latter helps to predict the future state of hosts and detect overloaded and underloaded hosts. We simulated the proposed algorithm using the CloudSim framework and compared it to the baselines and state-of-the-art algorithms. The results show that the proposed approach surpasses other methods in what concerns both performance and energy efficiency. [ABSTRACT FROM AUTHOR]

The emergence of new complex technologies, such as cloud and quantum computing, has a complicated computational structure, which can compromise the security of traditional cryptographic protocols. It has been noted that traditional security systems utilize non-quantum resistance factorization and Diffie–Hellman (DH) hard problems for their protection. Lattice-based cryptography appears to be a potential post-quantum substitute for the presently employed public-key cryptography. As a result, we have been motivated to focus on increasingly sophisticated and challenging lattice hard assumptions capable of withstanding new modern technologies with quantum resistance properties. In this paper, we devise a new lattice-based public-key encryption and signature schemes with smaller key sizes. The proposed scheme also resists the quantum attack because of the lattice small integer solution problem and its variant. The security claim has also been proved in a well-suited model for quantum attacks. The performance analysis shows that the presented schemes outperform the DH-type schemes and compete with similar lattice-based schemes in terms of storage, communication, key sizes, and computational overheads. [ABSTRACT FROM AUTHOR]

Abstract  Beowulf clusters are now deployed worldwide, chiefly in support of scientific computing. Beowulf clusters yield high computing performance, yet they also pose several challenges: (1) heat-induced hardware failure makes large scale commodity clusters fail quite frequently and (2) cost effectiveness of the Beowulf cluster is challenged by the fact that it lacks means of adapting its power state according to varying work load. This paper addresses these issues by developing a Power and Environment Awareness Module (PEAM) for a Beowulf cluster. The busty nature of computation load in an academic environment inspired the implementation and analysis of a fixed timeout Dynamic Power Management (DPM) policy. Today it is common that many Beowulf clusters in academic environment are composed of older, recycled nodes that may lack of out-of-band management technologies, thus Advanced Configuration and Power Interface (ACPI) and Wake-on-LAN (WOL) technology is exploited to control the power state of cluster nodes. A data center environment monitoring system that uses Wireless Sensor Networks (WSN) technology is developed and deployed to realize environment awareness of the cluster. Our PEAM module has been implemented on our cluster at Purdue University, reducing the operational cost and increasing the reliability of the cluster by reducing heat generation and optimizing workload distribution in an environment aware manner. [ABSTRACT FROM AUTHOR]

Abstract   Large-scale scientific computing applications frequently make use of closely-coupled distributed parallel components. The performance of such applications is therefore dependent on the component parts and their interaction at run-time. This paper describes a methodology for predictive performance modelling and evaluation of parallel applications composed of multiple interacting components. In this paper, the fundamental steps and required operations involved in the modelling and evaluation process are identified—including component decomposition, component model combination, MN communication modelling, dataflow analysis and overall performance evaluation. A case study is presented to illustrate the modelling process and the methodology is verified through experimental analysis. [ABSTRACT FROM AUTHOR]

Abstract  MPI collective communication operations to distribute or gather data are used for many parallel applications from scientific computing, but they may lead to scalability problems since their execution times increase with the number of participating processors. In this article, we show how the execution time of collective communication operations can be improved significantly by an internal restructuring based on orthogonal processor structures with two or more levels. The execution time of operations like MPI_Bcast() or MPI_Allgather() can be reduced by 40% and 70% on a dual Xeon cluster and a Beowulf cluster with single-processor nodes. But also on a Cray T3E a significant performance improvement can be obtained by a careful selection of the processor structure. The use of these optimized communication operations can reduce the execution time of data parallel implementations of complex application programs significantly without requiring any other change of the computation and communication structure. We present runtime functions for the modeling of two-phase realizations and verify that these runtime functions can predict the execution time both for communication operations in isolation and in the context of application programs. [ABSTRACT FROM AUTHOR]

A cloud data center provides various facilities such as storage, data accessibility, and running many specific applications on cloud resources. The unpredictable demand for service requests in cloud workloads affects the availability of resources during scheduling. It raises the issues of inaccurate workload prediction, lack of fulfillment in resource demands, load unbalancing, high power consumption due to heavy loads, and problems of under and overutilization of resources. Therefore, an efficient scheduling technique and an accurate forecasting model are needed to overcome these issues. Also, to deal with these challenges and provide optimal solutions, researchers must have a robust knowledge of cloud workloads, their types, issues, existing technologies, their advantages and disadvantages. However, previous research indicates limited systematic review studies exist for cloud workload applications with prediction-based scheduling techniques. Therefore, a survey is required that provides information related to cloud workload. To fulfill this requirement, the current study collects the related articles published in the past years. This paper is a systematic review study of prediction-based scheduling techniques that extract and evaluate data based on five criteria. It includes the datasets of different workload applications, resources, current prediction and scheduling techniques, and their related parameters. The survey is quite useful for academicians who want to select the problem and develop new techniques for issues related to cloud workload applications. It also gives an idea of existing approaches that are already implemented and employed. [ABSTRACT FROM AUTHOR]

Deep learning has been postulated as a solution for numerous problems in different branches of science. Given the resource-intensive nature of these models, they often need to be executed on specialized hardware such graphical processing units (GPUs) in a distributed manner. In the academic field, researchers get access to this kind of resources through High Performance Computing (HPC) clusters. This kind of infrastructures make the training of these models difficult due to their multi-user nature and limited user permission. In addition, different HPC clusters may possess different peculiarities that can entangle the research cycle (e.g., libraries dependencies). In this paper we develop a workflow and methodology for the distributed training of deep learning models in HPC clusters which provides researchers with a series of novel advantages. It relies on udocker as containerization tool and on Horovod as library for the distribution of the models across multiple GPUs. udocker does not need any special permission, allowing researchers to run the entire workflow without relying on any administrator. Horovod ensures the efficient distribution of the training independently of the deep learning framework used. Additionally, due to containerization and specific features of the workflow, it provides researchers with a cluster-agnostic way of running their models. The experiments carried out show that the workflow offers good scalability in the distributed training of the models and that it easily adapts to different clusters. [ABSTRACT FROM AUTHOR]