Journal of Parallel and Distributed Computing最新文献

There is an increasing need to provide real-time datasets for climate monitoring and applications. However, the current data products from all international groups have at least a month delay for data release. One reason for this delay is the long computing time of the global reconstruction algorithm (so-called mapping approach). To tackle this issue, this paper proposes a multi-level parallel computing model to improve the efficiency of data construction by parallelization of computation, reducing code branch prediction, optimizing data spatial locality, cache utilization, and other measures. This model has been applied to a mapping approach proposed by the Institute of Atmospheric Physics (IAP), one of the world's most widely used data products in the ocean and climate field. Compared with the traditional serial construction of MATLAB-based scheme on a single node, the speed of the construction after parallel optimizations is speeded up by ∼4.7 times. A large-scale parallel experiment of a long-term (∼1000 months) gridded dataset utilizing over 16,000 processor cores proves the model's scalability, improving ∼1200 times. In summary, this new model represents another example of the application of high-performance computing in oceanography and climatology.

Loops take up most of the time of computer programs, so optimizing them so that they run in the shortest time possible is a continuous task. However, this task is not negligible; on the contrary, it is an open area of research since many irregular loops are hard to parallelize. Generally, these loops have loop-carried (DOACROSS) dependencies and the appearance of dependencies could depend on the context. Many techniques have been studied to be able to parallelize these loops efficiently; however, for example in the OpenMP standard there is no efficient way to parallelize them. This article presents Speculative Task Execution (STE), a technique that enables the execution of OpenMP tasks in a speculative way to accelerate certain hot-code regions (such as loops) marked by OpenMP directives. It also presents a detailed analysis of the application of Hardware Transactional Memory (HTM) support for executing tasks speculatively and describes a careful evaluation of the implementation of STE using HTM on modern machines. In particular, we consider the scenario in which speculative tasks are generated by the OpenMP taskloop construct (Speculative Taskloop (STL)). As a result, it provides evidence to support several important claims about the performance of STE over HTM in modern processor architectures. Experimental results reveal that: (a) by implementing STL on top of HTM for hot-code regions, speed-ups of up to 5.39× can be obtained in IBM POWER8 and of up to 2.41× in Intel processors using 4 cores; and (b) STL-ROT, a variant of STL using rollback-only transactions (ROTs), achieves speed-ups of up to 17.70× in IBM POWER9 processor using 20 cores.

Stream Processing Systems (SPSs) dynamically process input events. Since the input is usually not a constant flow, presenting rate fluctuations, many works in the literature propose to dynamically replicate SPS operators, aiming at reducing the processing bottleneck induced by such fluctuations. However, these SPSs do not consider the problem of load balancing of the replicas or the cost involved in reconfiguring the system whenever the number of replicas changes. We present in this paper a predictive model which, based on input rate variation, execution time of operators, and queued events, dynamically defines the necessary current number of replicas of each operator. A predictor, composed of different models (i.e., mathematical and Machine Learning ones), predicts the input rate. We also propose a Storm-based SPS, named PA-SPS, which uses such a predictive model, not requiring reboot reconfiguration when the number of operators replica change. PA-SPS also implements a load balancer that distributes incoming events evenly among replicas of an operator. We have conducted experiments on Google Cloud Platform (GCP) for evaluation PA-SPS using real traffic traces of different applications and also compared it with Storm and other existing SPSs.

The Internet of Medical Things (IoMT) is a transformative fusion of medical sensors, equipment, and the Internet of Things, positioned to transform healthcare. However, security and privacy concerns hinder widespread IoMT adoption, intensified by the scarcity of high-quality datasets for developing effective security solutions. Addressing these challenges, we propose a novel framework for cyberattack detection in dynamic IoMT networks. This framework integrates Federated Learning with Meta-learning, employing a multi-phase architecture for identifying known attacks, and incorporates advanced clustering and biased classifiers to address zero-day attacks. The framework's deployment is adaptable to dynamic and diverse environments, utilizing an Infrastructure-as-a-Service (IaaS) model on the cloud and a Software-as-a-Service (SaaS) model on the fog end. To reflect real-world scenarios, we introduce a specialized IoMT dataset. Our experimental results indicate high accuracy and low misclassification rates, demonstrating the framework's capability in detecting cyber threats in complex IoMT environments. This approach shows significant promise in bolstering cybersecurity in advanced healthcare technologies.

Modern distributed systems are designed to manage overload conditions, by throttling the traffic in excess that cannot be served through overload control techniques. However, the adoption of large-scale NoSQL datastores make systems vulnerable to unbalanced overloads, where specific datastore nodes are overloaded because of hot-spot resources and hogs. In this paper, we propose DRACO, a novel overload control solution that is aware of data dependencies between the application and the datastore tiers. DRACO performs selective admission control of application requests, by only dropping the ones that map to resources on overloaded datastore nodes, while achieving high resource utilization on non-overloaded datastore nodes. We evaluate DRACO on two case studies with high availability and performance requirements, a virtualized IP Multimedia Subsystem and a distributed fileserver. Results show that the solution can achieve high performance and resource utilization even under extreme overload conditions, up to 100x the engineered capacity.

The rapid development of artificial intelligence and networking technologies has catalyzed the popularity of intelligent services based on deep learning in recent years, which in turn fosters the advancement of Web of Things (WoT). Big social data (BSD) plays an important role during the processing of intelligent services in WoT. However, intelligent BSD services are computationally intensive and require ultra-low latency. End or edge devices with limited computing power cannot realize the extremely low response latency of those services. Distributed inference of deep neural networks (DNNs) on various devices is considered a feasible solution by allocating the computing load of a DNN to several devices. In this work, an efficient model parallelism method that couples convolution layer (Conv) split with resource allocation is proposed. First, given a random computing resource allocation strategy, the Conv split decision is made through a mathematical analysis method to realize the parallel inference of convolutional neural networks (CNNs). Next, Deep Reinforcement Learning is used to get the optimal computing resource allocation strategy to maximize the resource utilization rate and minimize the CNN inference latency. Finally, simulation results show that our approach performs better than the baselines and is applicable for BSD services in WoT with a high workload.

Big data frameworks are widely deployed in supercomputers for analyzing large-scale datasets. Topological data processing is an emerging approach that focuses on analyzing the topological structures in high-dimensional scientific data. However, incorporating topological data processing into current big data frameworks presents three main challenges: (1) The frequent data exchange poses challenges to the traditional coarse-grained parallelism. (2) The spatial topology makes parallel programming harder using oversimplified MapReduce APIs. (3) The massive intermediate data and NUMA architecture hinder resource utilization and scalability on novel supercomputers and many-core processors.

In this paper, we present Topo, a generic distributed framework that enhances topological data processing on many-core supercomputers. Topo relies on three concepts. (1) It employs fine-grained parallelism, with awareness of topological structures in datasets, to support interactions among collaborative workers before each shuffle phase. (2) It provides intuitive APIs for topological data operations. (3) It implements efficient collective I/O and NUMA-aware dynamic task scheduling to improve multi-threading and load balancing. We evaluate Topo's performance on the Tianhe-3 supercomputer, which utilizes state-of-the-art ARM many-core processors. Experimental results of execution time show that compared to popular frameworks, Topo achieves an average speedup of 5.3× and 6.3×, with a maximum speedup of 8.4× and 20×, on HPC workloads and big data benchmarks, respectively. Topo further reduces total execution time on processing skewed datasets by 41%.

The folded hypercube is one of the hypercube variants and is of great significance in the study of interconnection networks. In a folded hypercube, information can be broadcast using efficient distributed algorithms. In the context of parallel computing, folded hypercube has been studied as a possible network topology as an alternative to the hypercube. The routing and wavelength assignment (RWA) problem is significant, since it improves the performance of wavelength-routed all-optical networks constructed using wavelength division multiplexing approach. Given the physical network topology, the aim of the RWA problem is to establish routes for the connection requests and assign the fewest possible wavelengths in accordance with the wavelength continuity and distinct wavelength constraints. This paper discusses the RWA problem in a linear array for the folded hypercube communication pattern by using the congestion technique.

Recent hardware-aware matrix-free algorithms for higher-order finite-element (FE) discretized matrix-vector multiplications reduce floating point operations and data access costs compared to traditional sparse matrix approaches. In this work, we address a critical gap in existing matrix-free implementations which are not well suited for the action of FE discretized matrices on very large number of vectors. In particular, we propose efficient matrix-free algorithms for evaluating FE discretized matrix-multivector products on both multi-node CPU and GPU architectures. To this end, we employ batched evaluation strategies, with the batchsize tailored to underlying hardware architectures, leading to better data locality and enabling further parallelization. On CPUs, we utilize even-odd decomposition, SIMD vectorization, and overlapping computation and communication strategies. On GPUs, we develop strategies to overlap compute with data movement for achieving efficient pipelining and reduced data accesses through the use of GPU-shared memory, constant memory and kernel fusion. Our implementation outperforms the baselines for Helmholtz operator action on 1024 vectors, achieving up to 1.4x improvement on one CPU node and up to 2.8x on one GPU node, while reaching up to 4.4x and 1.5x improvement on multiple nodes for CPUs (3072 cores) and GPUs (24 GPUs), respectively. We further benchmark the performance of the proposed implementation for solving a model eigenvalue problem for 1024 smallest eigenvalue-eigenvector pairs by employing the Chebyshev Filtered Subspace Iteration method, achieving up to 1.5x improvement on one CPU node and up to 2.2x on one GPU node while reaching up to 3.0x and 1.4x improvement on multi-node CPUs (3072 cores) and GPUs (24 GPUs), respectively.