Browsing by Author "Cheng, Yue"

Now showing 1 - 4 of 4
  • Thumbnail Image
    FedCaSe: Enhancing Federated Learning with Heterogeneity-aware Caching and Scheduling
    Khan, Redwan Ibne Seraj; Paul, Arnab K.; Jian, Xun (Steve); Cheng, Yue; Butt, Ali R. (ACM, 2024-11-20)
    Federated learning (FL) has emerged as a new paradigm of machine learning (ML) with the goal of collaborative learning on the vast pool of private data available across distributed edge devices. The focus of most existing works in FL systems has been on addressing the challenges of computation and communication heterogeneity inherent in training with edge devices. However, the crucial impact of I/O and the role of limited on-device storage has not been explored fully in FL context. Without policies to exploit the on-device storage for placement of client data samples, and schedule clients based on I/O benefits, FL training can lead to inefficiencies, such as increased training time and impacted accuracy convergence. In this paper, we propose FedCaSe, a framework for efficiently caching client samples in-situ on limited on-device storage and scheduling client participation. FedCaSe boosts the I/O performance by exploiting a unique characteristic— the experience, i.e., relative impact on overall performance, of data samples and clients. FedCaSe utilizes this information in adaptive caching policies for sample placement inside the limited memory of edge clients. The framework also exploits the experience information to orchestrate the future selection of clients. Our experiments with representative workloads and policies show that compared to the state of the art, FedCaSe improves the training time by 2.06× for accuracy convergence at the scale of thousands of clients.
  • Thumbnail Image
    MOANA: Modeling and Analyzing I/O Variability in Parallel System Experimental Design
    Cameron, Kirk W.; Anwar, Ali; Cheng, Yue; Xu, Li; Li, Bo; Ananth, Uday; Lux, Thomas; Hong, Yili; Watson, Layne T.; Butt, Ali R. (Department of Computer Science, Virginia Polytechnic Institute & State University, 2018-04-19)
    Exponential increases in complexity and scale make variability a growing threat to sustaining HPC performance at exascale. Performance variability in HPC I/O is common, acute, and formidable. We take the first step towards comprehensively studying linear and nonlinear approaches to modeling HPC I/O system variability. We create a modeling and analysis approach (MOANA) that predicts HPC I/O variability for thousands of software and hardware configurations on highly parallel shared-memory systems. Our findings indicate nonlinear approaches to I/O variability prediction are an order of magnitude more accurate than linear regression techniques. We demonstrate the use of MOANA to accurately predict the confidence intervals of unmeasured I/O system configurations for a given number of repeat runs – enabling users to quantitatively balance experiment duration with statistical confidence.
  • Thumbnail Image
    SHADE: Enable Fundamental Cacheability for Distributed Deep Learning Training
    Khan, Redwan; Yazdani, Ahmad; Fu, Yuqi; Paul, Arnab; Ji, Bo; Jian, Xun; Cheng, Yue; Butt, Ali (Usenix Association, 2023)
    Deep learning training (DLT) applications exhibit unique I/O workload behaviors that pose new challenges for storage system design. DLT is I/O intensive since data samples need to be fetched continuously from a remote storage. Accelerators such as GPUs have been extensively used to support these applications. As accelerators become more powerful and more data-hungry, the I/O performance lags behind. This creates a crucial performance bottleneck, especially in distributed DLT. At the same time, the exponentially growing dataset sizes make it impossible to store these datasets entirely in memory. While today’s DLT frameworks typically use a random sampling policy that treat all samples uniformly equally, recent findings indicate that not all samples are equally important and different data samples contribute differently towards improving the accuracy of a model. This observation creates an opportunity for DLT I/O optimizations by exploiting the data locality enabled by importance sampling. To this end, we design and implement SHADE, a new DLT-aware caching system that detects fine-grained importance variations at per-sample level and leverages the variance to make informed caching decisions for a distributed DLT job. SHADE adopts a novel, rank-based approach, which captures the relative importance of data samples across different minibatches. SHADE then dynamically updates the importance scores of all samples during training. With these techniques, SHADE manages to significantly improve the cache hit ratio of the DLT job, and thus, improves the job’s training performance. Evaluation with representative computer vision (CV) models shows that SHADE, with a small cache, improves the cache hit ratio by up to 4.5× compared to the LRU caching policy.
  • Thumbnail Image
    Workload-aware Efficient Storage Systems
    Cheng, Yue (Virginia Tech, 2017-08-07)
    The growing disparity in data storage and retrieval needs of modern applications is driving the proliferation of a wide variety of storage systems (e.g., key-value stores, cloud storage services, distributed filesystems, and flash cache, etc.). While extant storage systems are designed and tuned for a specific set of applications targeting a range of workload characteristics, they lack the flexibility in adapting to the ever-changing workload behaviors. Moreover, the complexities in implementing modern storage systems and adapting ever-changing storage requirements present unique opportunities and engineering challenges. In this dissertation, we design and develop a series of novel data management and storage systems solutions by applying a simple yet effective rule---workload awareness. We find that simple workload-aware data management strategies are effective in improving the efficiency of modern storage systems, sometimes by an order of magnitude. The first two works tackle the data management and storage space allocation issues at distributed and cloud storage level, while the third work focuses on low-level data management problems in the local storage system, which many high-level storage/data-intensive applications rely on. In the first part of this dissertation (Chapter 3), we propose and develop MBal, a high-performance in-memory object caching framework with adaptive multi-phase load balancing, which supports not only horizontal (scale-out) but vertical (scale-up) scalability as well. MBal is able to make efficient use of available resources in the cloud through its fine-grained, partitioned, lockless design. In the second part of this dissertation (Chapter 4 and Chapter5), we design and build CAST (Chapter 4), a Cloud Analytics Storage Tiering solution that cloud tenants can use to reduce monetary cost and improve performance of analytics workloads. The approach takes the first step towards providing storage tiering support for data analytics in the cloud. Furthermore, we propose a hybrid cloud object storage system (Chapter 5) that could effectively engage both the cloud service providers and cloud tenants via a novel dynamic pricing mechanism. In the third part of this dissertation (Chapter 6), targeting local storage, we explore offline algorithms for flash caching in terms of both hit ratio and flash lifespan. We design and implement a multi-stage heuristic by synthesizing several techniques that manage data at the granularity of a flash erasure unit (which we call a container) to approximate the offline optimal algorithm. In the fourth part of this dissertation (Chapter 7), we are focused on how to enable fast prototyping of efficient distributed key-value stores targeting a proxy-based layered architecture. In this work, we design and build {con}, a framework that significantly reduce the engineering effort required to build a full-fledged distributed key-value store. Our dissertation shows that simple workload-aware data management strategies can bring huge benefit in terms of both efficiency (i.e., performance, monetary cost, etc.) and flexibility (i.e., ease-of-use, ease-of-deployment, programmability, etc.). The principles of leveraging workload dynamicity and storage heterogeneity can be used to guide next-generation storage system software design, especially when being faced with new storage hardware technologies.