{"title":"SMore: Enhancing GPU Utilization in Deep Learning Clusters by Serverless-Based Co-Location Scheduling","authors":"Junhan Liu;Zinuo Cai;Yumou Liu;Hao Li;Zongpu Zhang;Ruhui Ma;Rajkumar Buyya","doi":"10.1109/TPDS.2025.3548320","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3548320","url":null,"abstract":"Deep learning (DL) clusters allow machine learning practitioners to submit their computation-intensive tasks, with GPUs accelerating their execution process. However, GPUs in current deep learning clusters are often under-utilized, which hampers the job performance and overall cluster throughput. It is urgent to improve GPU utilization, but existing works lack research on fine-grained allocation for GPU resources, as it typically allocates GPUs as indivisible units. Serverless computing reveals an opportunity to optimize utilization with fine-grained resource allocation methods, but it requires addressing three main challenges: co-location performance degradation, service level objectives guarantee of serverless functions, and cold start overhead. We propose <sc>SMore</small>, a framework based on serverless computing to optimize GPU resource utilization of DL clusters. <sc>SMore</small> dynamically predicts the possible co-location performance degradation and leverages a degradation-aware scheduling algorithm to ensure that the co-location decisions do not impact workload performance. It also dynamically preloads or offloads DL models by predicting the request numbers of the subsequent period to address the cold start issue. Through actual trace testing on the prototype of <sc>SMore</small>, we find that the average GPU utilization can be increased by 34% with degradation being controlled effectively.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"903-917"},"PeriodicalIF":5.6,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808946","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":"PimBeam: Efficient Regular Path Queries Over Graph Database Using Processing-in-Memory","authors":"Weihan Kong;Shengan Zheng;Yifan Hua;Ruoyan Ma;Yuheng Wen;Guifeng Wang;Cong Zhou;Linpeng Huang","doi":"10.1109/TPDS.2025.3547365","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3547365","url":null,"abstract":"Regular path queries (RPQs) in graph databases are bottlenecked by the memory wall. Emerging processing-in-memory (PIM) technologies offer a promising solution to dispatch and execute path matching tasks in parallel within PIM modules. We present an efficient PIM-based data management system tailored for RPQs and graph updates. Our solution, called PimBeam, facilitates efficient batch RPQs and graph updates by implementing a PIM-friendly dynamic graph partitioning algorithm. This algorithm effectively addresses graph skewness issues while maintaining graph locality with low overhead for handling RPQs. PimBeam streamlines label filtering queries by adding a filtering module on the PIM side and leveraging the parallelism of PIM. For the graph updates, PimBeam enhances processing efficiency by amortizing the host CPU's update overhead to PIM modules. Evaluation results of PimBeam indicate 3.59x speedup for RPQs and 29.33x speedup for graph update on average over the state-of-the-art traditional graph database.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"1042-1057"},"PeriodicalIF":5.6,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808948","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}
Keyun Cheng;Huancheng Puyang;Xiaolu Li;Patrick P. C. Lee;Yuchong Hu;Jie Li;Ting-Yi Wu
{"title":"Toward Load-Balanced Redundancy Transitioning for Erasure-Coded Storage","authors":"Keyun Cheng;Huancheng Puyang;Xiaolu Li;Patrick P. C. Lee;Yuchong Hu;Jie Li;Ting-Yi Wu","doi":"10.1109/TPDS.2025.3547872","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3547872","url":null,"abstract":"Redundancy transitioning enables erasure-coded storage to adapt to varying performance and reliability requirements by re-encoding data with new coding parameters on-the-fly. Existing studies focus on bandwidth-driven redundancy transitioning that reduces the transitioning bandwidth across storage nodes, yet the actual redundancy transitioning performance remains bottlenecked by the most loaded node. We present BART, a load-balanced redundancy transitioning scheme that aims to reduce the redundancy transitioning time via carefully scheduled parallelization. We show that finding an optimal load-balanced solution is difficult due to the large solution space. Given this challenge, BART decomposes the redundancy transitioning problem into multiple sub-problems and solves the sub-problems via efficient heuristics. We evaluate BART using both simulations for large-scale storage and HDFS prototype experiments on Alibaba Cloud. We show that BART significantly reduces the redundancy transitioning time compared with the bandwidth-driven approach.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"889-902"},"PeriodicalIF":5.6,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808836","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 Communication-Efficient Out-of-Core Graph Processing on the GPU","authors":"Qiange Wang;Xin Ai;Yongze Yan;Shufeng Gong;Yanfeng Zhang;Jing Chen;Ge Yu","doi":"10.1109/TPDS.2025.3547356","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3547356","url":null,"abstract":"The key performance bottleneck of large-scale graph processing on memory-limited GPUs is the host-GPU graph data transfer. Existing GPU-accelerated graph processing frameworks address this issue by managing the active subgraph transfer at runtime. Some frameworks adopt explicit transfer management approaches based on explicit memory copy with filter or compaction. In contrast, others adopt implicit transfer management approaches based on on-demand accesses with the zero-copy mechanism or unified virtual memory. Having made intensive analysis, we find that as the active vertices evolve, the performance of the two approaches varies in different workloads. Due to heavy redundant data transfers, high CPU compaction overhead, or low bandwidth utilization, adopting a single approach often results in suboptimal performance. Moreover, these methods lack effective cache management methods to address the irregular and sparse memory access pattern of graph processing. In this work, we propose a hybrid transfer management approach that takes the merits of both two transfer approaches at runtime. Moreover, we present an efficient vertex-centric graph caching framework that minimizes CPU-GPU communication by caching frequently accessed graph data at runtime. Based on these techniques, we present HytGraph, a GPU-accelerated graph processing framework, which is empowered by a set of effective task-scheduling optimizations to improve performance. Experiments on real-world and synthetic graphs show that HytGraph achieves average speedups of 2.5 ×, 5.0 ×, and 2.0 × compared to the state-of-the-art GPU-accelerated graph processing systems, Grus, Subway, and EMOGI, respectively.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"961-976"},"PeriodicalIF":5.6,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143809004","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":"IoT-Dedup: Device Relationship-Based IoT Data Deduplication Scheme","authors":"Yuan Gao;Liquan Chen;Jianchang Lai;Tianyi Wang;Xiaoming Wu;Shui Yu","doi":"10.1109/TPDS.2025.3544315","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3544315","url":null,"abstract":"The cyclical and continuous working characteristics of <italic>Internet of Things</i> (<italic>IoT</i>) devices make a large amount of the same or similar data, which can significantly consume storage space. To solve this problem, various secure data deduplication schemes have been proposed. However, existing deduplication schemes only perform deduplication based on data similarity, ignoring the internal connection among devices, making the existing schemes not directly applicable to parallel and distributed scenarios like IoT. Furthermore, since secure data deduplication leads to multiple users sharing same encryption key, which may lead to security issues. To this end, we propose a device relationship-based IoT data deduplication scheme that fully considers the IoT data characteristics and devices internal connections. Specifically, we propose a device relationship prediction approach, which can obtain device collaborative relationships by clustering the topology of their communication graph, and classifies the data types based on device relationships to achieve data deduplication with different security levels. Then, we design a similarity-preserving encryption algorithm, so that the security level of encryption key is determined by the data type, ensuring the security of the deduplicated data. In addition, two different data deduplication methods, identical deduplication and similar deduplication, have been designed to meet the privacy requirement of different data types, improving the efficiency of deduplication while ensuring data privacy as much as possible. We evaluate the performance of our scheme using five real datasets, and the results show that our scheme has favorable results in terms of both deduplication performance and computational cost.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"847-860"},"PeriodicalIF":5.6,"publicationDate":"2025-02-21","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808837","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":"Courier: A Unified Communication Agent to Support Concurrent Flow Scheduling in Cluster Computing","authors":"Zhaochen Zhang;Xu Zhang;Zhaoxiang Bao;Liang Wei;Chaohong Tan;Wanchun Dou;Guihai Chen;Chen Tian","doi":"10.1109/TPDS.2025.3543882","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3543882","url":null,"abstract":"As one of the pillars in cluster computing frameworks, coflow scheduling algorithms can effectively shorten the network transmission time of cluster computing jobs, thus reducing the job completion times and improving the execution performance. However, most of existing coflow scheduling algorithms failed to consider the influences of concurrent flows, which can degrade their performance under a massive number of concurrent flows. To fill the gap, we propose a unified communication agent named Courier to minimize the number of concurrent flows in cluster computing applications, which is compatible with the mainstream coflow scheduling approaches. To maintain the scheduling order given by the scheduling algorithms, Courier merges multiple flows between each pair of hosts into a unified flow, and determines its order based on that of origin flows. In addition, in order to adapt to various types of topologies, Courier introduces a control mechanism to adjust the number of flows while maintaining the scheduling order. Extensive large-scale trace-driven simulations have shown that Courier is compatible with existing scheduling algorithms, and outperforms the state-of-the-art approaches by about 30% under a variety of workloads and topologies.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"861-876"},"PeriodicalIF":5.6,"publicationDate":"2025-02-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143821656","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":"Spread+: Scalable Model Aggregation in Federated Learning With Non-IID Data","authors":"Huanghuang Liang;Xin Yang;Xiaoming Han;Boan Liu;Chuang Hu;Dan Wang;Xiaobo Zhou;Dazhao Cheng","doi":"10.1109/TPDS.2025.3539738","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3539738","url":null,"abstract":"Federated learning (FL) addresses privacy concerns by training models without sharing raw data, overcoming the limitations of traditional machine learning paradigms. However, the rise of smart applications has accentuated the heterogeneity in data and devices, which presents significant challenges for FL. In particular, data skewness among participants can compromise model accuracy, while diverse device capabilities lead to aggregation bottlenecks, causing severe model congestion. In this article, we introduce Spread+, a hierarchical system that enhances FL by organizing clients into clusters and delegating model aggregation to edge devices, thus mitigating these challenges. Spread+ leverages hedonic coalition formation game to optimize customer organization and adaptive algorithms to regulate aggregation intervals within and across clusters. Moreover, it refines the aggregation algorithm to boost model accuracy. Our experiments demonstrate that Spread+ significantly alleviates the central aggregation bottleneck and surpasses mainstream benchmarks, achieving performance improvements of 49.58% over FAVG and 22.78% over Ring-allreduce.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 4","pages":"701-716"},"PeriodicalIF":5.6,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143535516","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":"Libfork: Portable Continuation-Stealing With Stackless Coroutines","authors":"Conor J. Williams;James Elliott","doi":"10.1109/TPDS.2025.3543442","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3543442","url":null,"abstract":"Fully-strict fork-join parallelism is a powerful model for shared-memory programming due to its optimal time-scaling and strong bounds on memory scaling. The latter is rarely achieved due to the difficulty of implementing continuation-stealing in traditional High Performance Computing (HPC) languages – where it is often impossible without modifying the compiler or resorting to non-portable techniques. We demonstrate how stackless-coroutines (a new feature in C<b>++</b><inline-formula><tex-math>$bm {20}$</tex-math></inline-formula>) can enable fully-portable continuation stealing and present <i>libfork</i> a wait-free fine-grained parallelism library, combining coroutines with user-space, geometric segmented-stacks. We show our approach is able to achieve optimal time/memory scaling, both theoretically and empirically, across a variety of benchmarks. Compared to openMP (libomp), libfork is on average <inline-formula><tex-math>$7.2times$</tex-math></inline-formula> faster and consumes <inline-formula><tex-math>$10times$</tex-math></inline-formula> less memory. Similarly, compared to Intel's TBB, libfork is on average <inline-formula><tex-math>$2.7times$</tex-math></inline-formula> faster and consumes <inline-formula><tex-math>$6.2times$</tex-math></inline-formula> less memory. Additionally, we introduce non-uniform memory access (NUMA) optimizations for schedulers that demonstrate performance matching <i>busy-waiting</i> schedulers.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 5","pages":"877-888"},"PeriodicalIF":5.6,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143808951","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":"FedTune-SGM: A Stackelberg-Driven Personalized Federated Learning Strategy for Edge Networks","authors":"Neha Singh;Mainak Adhikari","doi":"10.1109/TPDS.2025.3543368","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3543368","url":null,"abstract":"Federated Learning (FL) has emerged as a prominent solution for distributed learning environments, enabling collaborative model training without centralized data collection. However, FL faces significant challenges such as data heterogeneity and resource-constraint edge devices for model training and analysis, leading to accuracy degradation and bias in model performance. To address these critical issues, we propose a novel FL strategy named FedTune-SGM, designed to optimize model training in decentralized settings. In this strategy, a cloud-based model is initially trained and fine-tuned on the edge devices with additional layers tailored to the specific data characteristics. This fine-tuning process effectively mitigates the impact of data heterogeneity, enhancing the robustness and generalization capability of the model. FedTune-SGM employs a strategic weighting mechanism that ensures a balanced and equitable contribution from participating edge devices to prevent dominant influences from resource-rich devices and promote a fairer and more accurate aggregated model. Additionally, the proposed strategy integrates a Stackelberg Game model to foster an interactive and dynamic cloud-edge setup that motivates edge devices to invest more effort in model training and ensures the effectiveness of resource-constraint edge devices. Extensive experiments conducted on three diverse datasets highlight the superior performance of the proposed FedTune-SGM strategy compared to state-of-the-art FL techniques in terms of accuracy and robustness while meeting the critical challenges of data heterogeneity and resource limitations in FL environments. Through these innovations, FedTune-SGM paves the way for more reliable and efficient distributed learning systems, unlocking the full potential of FL in practical applications.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 4","pages":"791-802"},"PeriodicalIF":5.6,"publicationDate":"2025-02-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553185","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":"A Tail Latency SLO Guaranteed Task Scheduling Scheme for User-Facing Services","authors":"Zhijun Wang;Huiyang Li;Lin Sun;Stoddard Rosenkrantz;Hao Che;Hong Jiang","doi":"10.1109/TPDS.2025.3542638","DOIUrl":"https://doi.org/10.1109/TPDS.2025.3542638","url":null,"abstract":"A primary design objective for user-facing services for cloud and edge computing is to maximize query throughput, while meeting query tail latency Service Level Objectives (SLOs) for individual queries. Unfortunately, the existing solutions fall short of achieving this design objective, which we argue, is largely attributed to the fact that they fail to take the query fanout explicitly into account. In this paper, we propose TailGuard based on a Tail-latency-SLO-and-Fanout-aware Earliest-Deadline-First Queuing policy (TF-EDFQ) for task queuing at individual task servers the query tasks are fanned out to. With the task pre-dequeuing time deadline for each task being derived based on both query tail latency SLO and query fanout, TailGuard takes an important first step towards achieving the design objective. A query admission control scheme is also developed to provide tail latency SLO guarantee in the presence of resource shortages. TailGuard is evaluated against First-In-First-Out (FIFO) task queuing, task PRIority Queuing (PRIQ) and Tail-latency-SLO-aware EDFQ (T-EDFQ) policies by both simulation and testing in the Amazon EC2 cloud. It is driven by three types of applications in the Tailbench benchmark suite, featuring web search, in-memory key-value store, and transactional database applications. The results demonstrate that TailGuard can significantly improve resource utilization (e.g., up to 80% compared to FIFO), while also meeting the targeted tail latency SLOs, as compared with the other three policies. TailGuard is also implemented and tested in a highly heterogeneous Sensing-<inline-formula><tex-math>$a$</tex-math></inline-formula>s-a-Service (SaS) testbed for a data sensing service, demonstrating performance gains of up to 33% . These results are consistent with both the simulation and Amazon EC2 results.","PeriodicalId":13257,"journal":{"name":"IEEE Transactions on Parallel and Distributed Systems","volume":"36 4","pages":"759-774"},"PeriodicalIF":5.6,"publicationDate":"2025-02-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143553350","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}