Future Generation Computer Systems-The International Journal of Escience最新文献

{"title":"pFL-SBPM: A communication-efficient personalized federated learning framework for resource-limited edge clients","authors":"Han Hu ,&nbsp;Wenli Du ,&nbsp;Yuqiang Li ,&nbsp;Yue Wang","doi":"10.1016/j.future.2025.107849","DOIUrl":"10.1016/j.future.2025.107849","url":null,"abstract":"<div><div>Federated learning has attracted widespread attention due to its privacy-preserving characteristic. However, in real-world scenarios, the heterogeneity of decentralized data and the limited communication resources of clients pose great challenges to the deployment of federated training. Although existing works have made great strides in dealing with heterogeneous data or compressing communication, they struggle to strike a balance between model accuracy and communication cost. To address the above issues, this paper proposes a novel federated learning framework called pFL-SBPM, which achieves communication-efficient <strong>p</strong>ersonalized <strong>F</strong>ederated <strong>L</strong>earning through <strong>S</strong>tochastic <strong>B</strong>inary <strong>P</strong>robability <strong>M</strong>asks. Specifically, we utilize probability mask optimization instead of conventional weight training, where clients obtain personalized sparse subnetworks adapted to local task requirements by cooperative optimization of probability masks in a randomly weighted network. We develop an uplink communication strategy based on stochastic binary masks and a downlink communication strategy based on binary encoding and decoding, which achieves enhanced privacy protection while dramatically reducing the communication cost. Furthermore, to effectively handle heterogeneous data while mitigating the negative impact of the introduction of stochasticity on the stability of federated training, we carefully design a soft-threshold based selective updating strategy for probability masks. The experimental results show the significant superiority and competitiveness of pFL-SBPM compared to existing baseline and state-of-the-art methods in terms of inference accuracy, communication cost, computational cost and model size.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"171 ","pages":"Article 107849"},"PeriodicalIF":6.2,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143856017","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"The role of Large Language Models in addressing IoT challenges: A systematic literature review","authors":"Gabriele De Vito,&nbsp;Fabio Palomba,&nbsp;Filomena Ferrucci","doi":"10.1016/j.future.2025.107829","DOIUrl":"10.1016/j.future.2025.107829","url":null,"abstract":"<div><div>The Internet of Things (IoT) has revolutionized various sectors by enabling devices to communicate and interact seamlessly. However, developing IoT applications has data management, security, and interoperability challenges. Large Language Models (LLMs) have shown promise in addressing these challenges due to their advanced language processing capabilities. This Systematic Literature Review assesses the role of LLMs in addressing IoT challenges, exploring the strategies, hardware, and software configurations used, and identifying directions for future research. We extensively searched databases like Scopus, IEEE Xplore, and ACM Digital Library, initially screening 1,419 studies and identifying an additional 1,167 through snowballing, ultimately focusing on 55 relevant papers. The findings reveal LLMs’ potential to address key IoT challenges such as security and scalability. However, they also highlight significant obstacles, including high computational demands and the complexities of training and tuning these models. Future research should aim to develop methods to reduce the computational requirements of LLMs, improve training datasets, simplify implementation processes, and explore the ethical and privacy implications of using LLMs in IoT applications.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"171 ","pages":"Article 107829"},"PeriodicalIF":6.2,"publicationDate":"2025-04-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143838900","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"DyCE: Dynamically Configurable Exiting for deep learning compression and real-time scaling","authors":"Qingyuan Wang ,&nbsp;Barry Cardiff ,&nbsp;Antoine Frappé ,&nbsp;Benoit Larras ,&nbsp;Deepu John","doi":"10.1016/j.future.2025.107837","DOIUrl":"10.1016/j.future.2025.107837","url":null,"abstract":"<div><div>Conventional deep learning (DL) model compression methods affect all input samples equally. However, as samples vary in difficulty, a dynamic model that adapts computation based on sample complexity offers a novel perspective for compression and scaling. Despite this potential, existing dynamic techniques are typically monolithic and have model-specific implementations, limiting their generalizability as broad compression and scaling methods. Additionally, most deployed DL systems are fixed, and unable to adjust once deployed. This paper introduces DyCE, a dynamically configurable system that can adjust the performance-complexity trade-off of a DL model at runtime without needing re-initialization or re-deployment. DyCE achieves this by adding exit networks to intermediate layers, thus allowing early termination if results are acceptable. DyCE also decouples the design of exit networks from the base model itself, enabling its easy adaptation to new base models. We also propose methods for generating optimized configurations and determining exit network types and positions for dynamic trade-offs. By enabling simple configuration switching, DyCE enables fine-grained performance-complexity tuning in real-time. We demonstrate the effectiveness of DyCE through image classification tasks using deep convolutional neural networks (CNNs). DyCE significantly reduces computational complexity by 26.2% for ResNet<span><math><msub><mrow></mrow><mrow><mi>152</mi></mrow></msub></math></span>, 26.6% for ConvNextv2<span><math><msub><mrow></mrow><mrow><mi>tiny</mi></mrow></msub></math></span> and 32.0% for DaViT<span><math><msub><mrow></mrow><mrow><mi>base</mi></mrow></msub></math></span> on ImageNet validation set, with accuracy reductions of less than 0.5%.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"171 ","pages":"Article 107837"},"PeriodicalIF":6.2,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143821298","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards dynamic virtual machine placement based on safety parameters and resource utilization fluctuation for energy savings and QoS improvement in cloud computing","authors":"Dan Wang ,&nbsp;Jinjiang Wang ,&nbsp;Xize Liu ,&nbsp;Junyang Yu ,&nbsp;Hangyu Gu ,&nbsp;Congyang Wang ,&nbsp;Jinghan Liu ,&nbsp;Yanhao Zhang","doi":"10.1016/j.future.2025.107853","DOIUrl":"10.1016/j.future.2025.107853","url":null,"abstract":"<div><div>The majority of studies regard virtual machine placement (VMP) as a multi-dimensional bin packing problem. The most common solution is to place as many virtual machines (VMs) on physical machines (PMs) as possible in order to improve the overall resource utilization of the cloud data centers (CDCs). However, it brings some obstacles for the working performance of VMs and the quality of service (QoS) for CDCs, such as (i) performance degradation between running VMs since resources are contested and (ii) resource wastage since single-dimensional resources fail to allocate resources for a new VM instance. In addition, we find that if we do not capture resource request fluctuation for tasks running on VMs, it will increase the probability of underloading or overloading the PM, resulting in VM migration with more performance loss and service-level agreement (SLA) violation (SLAV).</div><div>In order to solve the above problems and further realize the objectives of energy savings, QoS guarantee, we introduce an enhancing QoS dynamic virtual machine placement (ESVMP) mechanism. It is based on balance deviation factor, safety factor, and fluctuation factor indicators, which optimize PM resource utilization while deploying VMs on PMs that have balanced and stable resource utilization and the ability to guarantee VMs’ working performance so as to guarantee the QoS of the CDC. In addition, to further reduce energy consumption, the ESVMP algorithm leverages energy-efficiency indicators. The results of extensive experiments conducted on the CloudSim simulator show that under the PlanetLab workload, the ESVMP approach is able to reduce the energy consumption, number of migrations, SLAV, and ESV of CDCs by 2.8%, 61.5%, 98.5%, and 98.7%, respectively, on average, compared with the LBVMP approach; and under the Bitbrains workload, the ESVMP approach can reduce the number of migrations, SLAV, and ESV of CDCs by 60.9%, 89.7%, and 89.8% on average, respectively, compared with the LBVMP approach.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"171 ","pages":"Article 107853"},"PeriodicalIF":6.2,"publicationDate":"2025-04-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143856016","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CASR: Optimizing cold start and resources utilization in serverless computing","authors":"Yu Chen ,&nbsp;Bo Liu ,&nbsp;Weiwei Lin ,&nbsp;Yulin Guo ,&nbsp;Zhiping Peng","doi":"10.1016/j.future.2025.107851","DOIUrl":"10.1016/j.future.2025.107851","url":null,"abstract":"<div><div>Serverless computing, also known as Functions as a Service (FaaS), is an emerging cloud deployment paradigm that offers advantages such as pay-as-you-go pricing and automatic scaling. Functions often suffer from cold starts delays due to the overhead of initializing code and data dependencies before execution. Retaining containers in memory for a period after execution can reduce cold start latency. However, existing application-layer solutions overlook the effect of cold start overhead on the availability of containers, resulting in suboptimal balance between cold start latency and memory resource utilization. Furthermore, these strategies typically overlook the optimization of overall cold start overhead, which is essential for enhancing system efficiency. To address these challenges, we propose the Cache-Based Adaptive Scheduler for Serverless Runtime (CASR), an adaptive strategy for managing container runtime configurations. CASR effectively balances cold start latency and memory utilization while reducing overall cold start overhead. Specifically, we introduce a serverless cache (S-Cache) that leverages the equivalence between caching problems and container keep-alive strategies to mitigate cold starts. Additionally, we develop a deep reinforcement learning model, based on the proximal policy optimization algorithm, to enable the automatic scaling of the S-Cache queue, allowing adaptation to dynamic cloud workloads and enhancing memory resource utilization. Extensive simulations on an Azure dataset show that CASR reduces cold starts by 38.75%, improves memory resource utilization by 46.73%, and decreases cold start overhead by 48.53% compared to existing container keep-alive strategies in serverless platforms under common workloads.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"170 ","pages":"Article 107851"},"PeriodicalIF":6.2,"publicationDate":"2025-04-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143817084","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Analyzing the performance portability of SYCL across CPUs, GPUs, and hybrid systems with SW sequence alignment","authors":"Manuel Costanzo ,&nbsp;Enzo Rucci ,&nbsp;Carlos García-Sánchez ,&nbsp;Marcelo Naiouf ,&nbsp;Manuel Prieto-Matías","doi":"10.1016/j.future.2025.107838","DOIUrl":"10.1016/j.future.2025.107838","url":null,"abstract":"<div><div>The high-performance computing (HPC) landscape is undergoing rapid transformation, with an increasing emphasis on energy-efficient and heterogeneous computing environments. This comprehensive study extends our previous research on SYCL’s performance portability by evaluating its effectiveness across a broader spectrum of computing architectures, including CPUs, GPUs, and hybrid CPU–GPU configurations from NVIDIA, Intel, and AMD. Our analysis covers single-GPU, multi-GPU, single-CPU, and CPU–GPU hybrid setups, using two common, bioinformatic applications as a case study. The results demonstrate SYCL’s versatility across different architectures, maintaining comparable performance to CUDA on NVIDIA GPUs while achieving similar architectural efficiency rates on AMD and Intel GPUs in the majority of cases tested. SYCL also demonstrated remarkable versatility and effectiveness across CPUs from various manufacturers, including the latest hybrid architectures from Intel. Although SYCL showed excellent functional portability in hybrid CPU–GPU configurations, performance varied significantly based on specific hardware combinations. Some performance limitations were identified in multi-GPU and CPU–GPU configurations, primarily attributed to workload distribution strategies rather than SYCL-specific constraints. These findings position SYCL as a promising unified programming model for heterogeneous computing environments, particularly for bioinformatic applications.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"170 ","pages":"Article 107838"},"PeriodicalIF":6.2,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808733","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Portability efficiency approach for calculating performance portability","authors":"Ami Marowka","doi":"10.1016/j.future.2025.107826","DOIUrl":"10.1016/j.future.2025.107826","url":null,"abstract":"<div><div>The emergence of heterogeneity in high-performance computing, which harnesses under one integrated system several platforms of different architectures, also led to the development of innovative cross-platform programming models. Along with the expectation that these models will yield computationally intensive performance, there is demand for them to provide a reasonable degree of performance portability. Therefore, new tools and metrics are being developed to measure and calculate the level of performance portability of applications and programming models.</div><div>The ultimate measure of performance portability is performance efficiency. Performance efficiency refers to the achieved performance as a fraction of some peak theoretical or practical baseline performance. <em>Application efficiency</em> approaches are the most popular and attractive performance efficiency measures among researchers because they are simple to measure and calculate. Unfortunately, the way they are used yields results that do not make sense, while violating one of the basic criteria that defines and characterizes the performance portability metrics.</div><div>In this paper, we demonstrate how researchers currently use application efficiency to calculate the performance portability of applications and explain why this method deviates from its original definition. Then, we show why the obtained results do not make sense and propose practical solutions that satisfy the definition and criteria of performance portability metrics.</div><div>Finally, we present a new performance efficiency approach called <em>portability efficiency</em>, which is immune to the shortcomings of application efficiency and substantially improves the aspect of portability when calculating <em>performance portability</em>.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"170 ","pages":"Article 107826"},"PeriodicalIF":6.2,"publicationDate":"2025-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143800125","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Recognition of Best Paper, Outstanding Editors, and Reviewers for Future Generation Computer Systems in 2024","authors":"Michela Taufer","doi":"10.1016/j.future.2025.107854","DOIUrl":"10.1016/j.future.2025.107854","url":null,"abstract":"","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"171 ","pages":"Article 107854"},"PeriodicalIF":6.2,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143864739","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Evaluation of Juliana Tool: A translator for Julia’s CUDA.jl code into KernelAbstraction.jl","authors":"Enrique de la Calle,&nbsp;Carlos García","doi":"10.1016/j.future.2025.107813","DOIUrl":"10.1016/j.future.2025.107813","url":null,"abstract":"<div><div>Julia is a high-level language that supports the execution of parallel code through various packages. CUDA.jl is widely used for developing GPU-accelerated code in Julia and is integrated into many libraries and programs. In this paper, we present Juliana, a novel tool that automatically translates Julia code utilizing the CUDA.jl package into an abstract, multi-backend representation powered by the KernelAbstractions.jl package. To evaluate the tool’s viability and performance, we analyzed four Julia projects: Rodinia, miniBUDE, BabelStream, and Oceananigans.jl. The performance overhead of this approach was found to be relatively low (under 7% for the Rodinia suite), with performance portability metrics showing results nearly identical to the native implementations. By running the same code across multiple KernelAbstractions’ backends, we successfully executed these translated projects on GPUs from vendors such as NVIDIA, Intel, AMD, and Apple. This ensured compatibility across these platforms and enabled first-time execution on some devices.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"171 ","pages":"Article 107813"},"PeriodicalIF":6.2,"publicationDate":"2025-04-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143833915","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"perfCorrelate: Performance variability correlation framework","authors":"Panagiotis Giannakopoulos ,&nbsp;Bart van Knippenberg ,&nbsp;Kishor Chandra Joshi ,&nbsp;Nicola Calabretta ,&nbsp;George Exarchakos","doi":"10.1016/j.future.2025.107827","DOIUrl":"10.1016/j.future.2025.107827","url":null,"abstract":"<div><div>Edge computing is a promising technology for deploying time-sensitive and privacy-sensitive applications closer to the premises of users. However, it is crucial to identify the sources of performance variability caused by application co-location to meet user requirements effectively. Monitoring systems typically expose hundreds of metrics, making comprehensive analysis challenging. As a result, researchers often rely on a small, arbitrarily selected subset of metrics for tasks such as building performance predictors. In this paper, we examine how the available monitoring metrics are correlated with Round Trip Time (RTT) fluctuations and suggest directions for building performance models. Our experiments focus on a Single Particle Analysis (SPA) applications for an electron microscopy use case, deployed in a Kubernetes environment and monitored by Prometheus. We demonstrate that while a subset of monitoring metrics consistently correlates with performance, the specific metrics in this subset can vary due to dynamic application co-locations and observation windows. Consequently, the optimal number of metrics and the choice of machine learning model needed to accurately capture performance variability vary between different scenarios (co-location and cluster nodes). These differences directly impact the effectiveness of scheduling decisions in resource clusters, which depend on performance predictors. Our work presents a method to systematically identify the most relevant monitoring metrics to changes in RTT and determining the most representative observation window, ensuring a more generalizable understanding of the performance of the application throughout its lifecycle.</div></div>","PeriodicalId":55132,"journal":{"name":"Future Generation Computer Systems-The International Journal of Escience","volume":"170 ","pages":"Article 107827"},"PeriodicalIF":6.2,"publicationDate":"2025-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808732","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}