Journal of Parallel and Distributed Computing最新文献

We propose a novel decentralized federated learning framework called B2DFL. It decomposes the aggregation process of vanilla FL into layered and serialized sub-aggregation processes and offloads the communication and computation from a single point to distributed nodes, thus addressing the single point of failure issue in centralized FL. The decentralization of B2DFL is based on the Butterfly, a distributed network topology, to organize and orchestrate the order and rules of node aggregation. Additionally, to mitigate potential risks such as dropouts or tampering, we leverage the blockchain and IPFS systems. Specifically, after each node completes its computation (including training and aggregation), it generates a hash value of the results as proof. We maintain a Tamper-evident Data Structure (TDS) on the blockchain, which records these proofs to ensure tamper-proofing and fast verification. To reduce the storage burden on the blockchain and improve throughput, we store the aggregated results on IPFS, a system that enables quick data location through hash values of data, for data backup. We also design a node replacement mechanism for quick dropout handling. We conduct a comprehensive performance evaluation and experimental results demonstrate that B2DFL presents a significant performance improvement while achieving privacy and decentralization.

Fortran's prominence in scientific computing requires strategies to ensure both that legacy codes are efficient on high-performance computing systems, and that the language remains attractive for the development of new high-performance codes. Coarray Fortran (CAF), part of the Fortran 2008 standard introduced for parallel programming, facilitates distributed memory parallelism with a syntax familiar to Fortran programmers, simplifying the transition from single-processor to multi-processor coding. This research focuses on innovating and refining a parallel programming methodology that fuses the strengths of Intel Coarray Fortran, Nvidia CUDA Fortran, and OpenMP for distributed memory parallelism, high-speed GPU acceleration and shared memory parallelism respectively. We consider the management of pageable and pinned memory, CPU-GPU affinity in NUMA multiprocessors, and robust compiler interfacing with speed optimisation. We demonstrate our method through its application to a parallelised Poisson solver and compare the methodology, implementation, and scaling performance to that of the Message Passing Interface (MPI), finding CAF offers similar speeds with easier implementation. For new codes, this approach offers a faster route to optimised parallel computing. For legacy codes, it eases the transition to parallel computing, allowing their transformation into scalable, high-performance computing applications without the need for extensive re-design or additional syntax.

Distributed computing, such as cloud computing, provides promising platforms for orchestrating scientific workflows' tasks based on their sequences and dependencies. Workflow scheduling plays an important role in optimizing concerned objectives for distributed computing, such as minimizing the makespan and cost. Many researchers have focused on optimizing a specific single workflow with multiple objectives. Currently, there are few studies on multi-workflow scheduling, with most research focusing on objectives such as cost and makespan. However, multi-workflow scheduling requires the design of specific objectives that reflect the unique characteristics of multiple workflows. On the other hand, clustering-based approaches have garnered significant attention in the field of workflow scheduling over distributed computing resources due to their advantage in reducing data communication among tasks. Despite this, the effectiveness of clustering-based algorithms has not been extensively studied and validated in the context of multi-objective multi-workflow scheduling models. Motivated by these factors, we propose an approach for multiple workflows' multi-objective optimization (MOO), considering the new defined metric, fairness. We first mathematically formulate the fairness and define a fairness-involved MOO model. Then, we propose an advanced clustering-based resource optimization strategy in multiple workflow runs. Experimental results show that the proposed approach performs better than the compared algorithms without significant compromise of the overall makespan and cost as well as individual fairness, which can guide the simulation workflow scheduling on clouds.

Graphs model several real-world phenomena. With the growth of unstructured and semi-structured data, parallelization of graph algorithms is inevitable. Unfortunately, due to inherent irregularity of computation, memory access, and communication, graph algorithms are traditionally challenging to parallelize. To tame this challenge, several libraries, frameworks, and domain-specific languages (DSLs) have been proposed to reduce the parallel programming burden of the users, who are often domain experts. However, existing frameworks to model graph algorithms typically target a single architecture. In this paper, we present a graph DSL, named StarPlat, that allows programmers to specify graph algorithms in a high-level format, but generates code for three different backends from the same algorithmic specification. In particular, the DSL compiler generates OpenMP for multi-core systems, MPI for distributed systems, and CUDA for many-core GPUs. Since these three are completely different parallel programming paradigms, binding them together under the same language is challenging. We share our experience with the language design. Central to our compiler is an intermediate representation which allows a common representation of the high-level program, from which individual backend code generations begin. We demonstrate the expressiveness of StarPlat by specifying four graph algorithms: betweenness centrality computation, page rank computation, single-source shortest paths, and triangle counting. Using a suite of ten large graphs, we illustrate the effectiveness of our approach by comparing the performance of the generated codes with that obtained with hand-crafted library codes. We find that the generated code is competitive to library-based codes in many cases. More importantly, we show the feasibility to generate efficient codes for different target architectures from the same algorithmic specification of graph algorithms.

Clustering is a pivotal primitive for unsupervised learning and data analysis. A popular variant is the $(k,\ell )$-clustering problem, where, given a pointset P from a metric space, one must determine a subset S of k centers minimizing the sum of the ℓ-th powers of the distances of points in P from their closest centers. This formulation covers the well-studied k-median ($\ell =1$) and k-means ($\ell =2$) clustering problems. A more general variant, introduced to deal with noisy pointsets, features a further parameter z and allows up to z points of P (outliers) to be disregarded when computing the sum. We present a distributed coreset-based 3-round approximation algorithm for the $(k,\ell )$-clustering problem with z outliers, using MapReduce as a computational model. An important feature of our algorithm is that it obliviously adapts to the intrinsic complexity of the dataset, captured by its doubling dimension D. Remarkably, for $D=O\left(1\right)$, our algorithm requires sublinear local memory per reducer, and yields a solution whose approximation ratio is an additive term $O\left(\gamma \right)$ away from the one achievable by the best known sequential (possibly bicriteria) algorithm, where γ can be made arbitrarily small. To the best of our knowledge, no previous distributed approaches were able to attain similar quality-performance tradeoffs for metrics with constant doubling dimension.

Recently, accelerator-based compression/decompression was proposed to hide the storage latency of high-performance computing (HPC) applications that generate/ingest large data that cannot fit a node's memory. In this work, such a scheme has been implemented using a novel FPGA-based lossy compression/decompression scheme that has very low-latency. The proposed scheme completely overlaps the movement of the application's data with its compute kernels on the CPU with minimal impact on these kernels. Experiments showed that it can yield performance levels on-par with utilizing memory-only storage buffers, even though data is actually stored on disk. Experiments also showed that compared to CPU- and GPU-based compression frameworks, it achieves better performance levels at a fraction of the power consumption.

Background

Hospitals and medical facilities are increasingly concerned about network security and patient data privacy as the Internet of Medical Things (IoMT) infrastructures continue to develop. Researchers have studied customized network security frameworks and cyberattack detection tools driven by Artificial Intelligence (AI) to counter different types of attacks, such as spoofing, data alteration, and botnet attacks. However, carrying out routine IoMT services and tasks during an under-attack scenario is challenging. Machine Learning has been extensively suggested for detecting cyberattacks in IoMT and IoT infrastructures. However, the conventional centralized approach in ML cannot effectively detect newly emerging attacks without compromising patient data privacy and network flow data confidentiality.

Aim

This study discusses a Federated Bayesian Optimization XGBoost framework that employs multimodal sensory signals from patient vital signs and network flow data to detect attack patterns and malicious network traffic in IoMT infrastructure while ensuring data privacy and detecting previously unknown attacks.

Methodology

The proposed model employs a Federated Bayesian Optimisation XGBoost approach, which allows us to search the parameter space quickly and find an optimal solution from each local server while aggregating the model parameters from each local server to the centralised server. The XGBoost algorithm generates a new tree by taking into account the previously estimated value for the tree's input data and then optimizing the prediction gain. This study used a dataset with 44 attributes and 16 318 instances. During the preprocessing phase, 10 features were dropped, and the remaining 34 features were used to evaluate the network flows and biometric data (patient vital signs).

Results

The performance evaluation reveals that the proposed model predicts data alteration, malware, and spoofing attacks in patients' vital signs and network flow data with a prediction accuracy of 0.96. The results obtained from the experiment demonstrate that both the centralized and federated models are synchronized, with the latter occasionally being slightly reduced.

Conclusion

The findings indicate that the suggested model can be incorporated into the IoMT domain to detect malicious patterns while maintaining data privacy and confidentiality efficiently.

A trajectory privacy protection method using cached candidate result sets (TPP-CCRS) is proposed for the user trajectory privacy leakage problem. First, the user's area is divided into a grid to lock the user's trajectory range, and a cache area is set on the user's mobile side to cache the candidate result sets queried from the user's area. Second, a security center is deployed to register users securely and assign public and private keys for verifying location information. The same user's location information is randomly divided into M copies and sent to multi-anonymizers. Then, the random concurrent k-anonymization mechanism with multi-anonymizers is used to concurrently k-anonymize M copies of location information. Finally, the prefix tree is added on the location-based service (LBS) server side, and the location information is encrypted using the clustered data fusion privacy protection algorithm. The optimal binary tree algorithm queries user interest points. Security analysis and experimental verification show that the TPP-CCRS can effectively protect user trajectory privacy and improve location information query efficiency.