Search for: All records

Recent advances in Deep Neural Networks (DNNs) have demonstrated a promising potential in predicting the temporal and spatial proximity of time evolutionary data. In this paper, we have developed an effective (de)compression framework called TEZIP that can support dynamic lossy and lossless compression of time evolutionary image frames with high compression ratio and speed. TEZIP first trains a Recurrent Neural Network called PredNet to predict future image frames based on base frames, and then derives the resulting differences between the predicted frames and the actual frames as more compressible delta frames. Next we equip TEZIP with techniques that can exploit spatial locality for the encoding of delta frames and apply lossless compressors on the resulting frames. Furthermore, we introduce window-based prediction algorithms and dynamically pinpoint the trade-off between the window size and the relative errors of predicted frames. Finally, we have conducted an extensive set of tests to evaluate TEZIP. Our experimental results show that, in terms of compression ratio, TEZIP outperforms existing lossless compressors such as x265 by up to 3.2x and lossy compressors such as SZ by up to 3.3x. 

With the emergence of versatile storage systems, multi-level checkpointing (MLC) has become a common approach to gain efficiency. However, multi-level checkpoint/restart can cause enormous I/O traffic on HPC systems. To use multilevel checkpointing efficiently, it is important to optimize checkpoint/restart configurations. Current approaches, namely modeling and simulation, are either inaccurate or slow in determining the optimal configuration for a large scale system. In this paper, we show that machine learning models can be used in combination with accurate simulation to determine the optimal checkpoint configurations. We also demonstrate that more advanced techniques such as neural networks can further improve the performance in optimizing checkpoint configurations. 

On large-scale high performance computing (HPC) systems, applications are provisioned with aggregated resources to meet their peak demands for brief periods. This results in resource underutilization because application requirements vary a lot during execution. This problem is particularly pronounced for deep learning applications that are running on leadership HPC systems with a large pool of burst buffers in the form of flash or non-volatile memory (NVM) devices. In this paper, we examine the I/O patterns of deep neural networks and reveal their critical need of loading many small samples randomly for successful training. We have designed a specialized Deep Learning File System (DLFS) that provides a thin set of APIs. Particularly, we design the metadata management of DLFS through an in-memory tree-based sample directory and its file services through the user-level SPDK protocol that can disaggregate the capabilities of NVM Express (NVMe) devices to parallel training tasks. Our experimental results show that DLFS can dramatically improve the throughput of training for deep neural networks on NVMe over Fabric, compared with the kernel-based Ext4 file system. Furthermore, DLFS achieves efficient user-level storage disaggregation with very little CPU utilization. 

Parallel File Systems (PFSs) are frequently deployed on leadership High Performance Computing (HPC) systems to ensure efficient I/O, persistent storage and scalable performance. Emerging Deep Learning (DL) applications incur new I/O and storage requirements to HPC systems with batched input of small random files. This mandates PFSs to have commensurate features that can meet the needs of DL applications. BeeGFS is a recently emerging PFS that has grabbed the attention of the research and industry world because of its performance, scalability and ease of use. While emphasizing a systematic performance analysis of BeeGFS, in this paper, we present the architectural and system features of BeeGFS, and perform an experimental evaluation using cutting-edge I/O, Metadata and DL application benchmarks. Particularly, we have utilized AlexNet and ResNet-50 models for the classification of ImageNet dataset using the Livermore Big Artificial Neural Network Toolkit (LBANN), and ImageNet data reader pipeline atop TensorFlow and Horovod. Through extensive performance characterization of BeeGFS, our study provides a useful documentation on how to leverage BeeGFS for the emerging DL applications. 

Apache Spark is a popular cluster computing framework for iterative analytics workloads due to its use of Resilient Distributed Datasets (RDDs) to cache data for in-memory processing. We have revealed that the performance of Spark RDD cache can be severely limited if its capacity falls short to the needs of the workloads. In this paper, we have explored different memory hybridization strategies to leverage emergent Non-Volatile Memory (NVM) devices for Spark's RDD cache. We have found that a simple layered hybridization approach does not offer an effective solution. Therefore, we have designed a flat hybridization scheme to leverage NVM for caching RDD blocks, along with several architectural optimizations such as dynamic memory allocation for block unrolling, asynchronous migration with preemption, and opportunistic eviction to disk. We have performed an extensive set of experiments to evaluate the performance of our proposed flat hybridization strategy and found it to be robust in handling different system and NVM characteristics. Our proposed approach uses DRAM for a fraction of the hybrid memory system and yet manages to keep the increase in execution time to be within 10% on average. Moreover, our opportunistic eviction of blocks to disk improves performance by up to 7.5% when utilized alongside the current mechanism. 

State-of-the-art synchronous graph processing frameworks face both inefficiency and imbalance issues that cause their performance to be suboptimal. These issues include the inefficiency of communication and the imbalanced graph computation/communication costs in an iteration. We propose to replace their conventional two-sided communication model with the one-sided counterpart. Accordingly, we design SHMEMGraph, an efficient and balanced graph processing framework that is formulated across a global memory space and takes advantage of the flexibility and efficiency of one-sided communication for graph processing. Through an efficient one-sided communication channel, SHMEMGraph utilizes the high-performance operations with RDMA while minimizing the resource contention within a computer node. In addition, SHMEMGraph synthesizes a number of optimizations to address both computation imbalance and communication imbalance. By using a graph of 1 billion edges, our evaluation shows that compared to the state-of-the-art Gemini framework, SHMEMGraph achieves an average improvement of 35.5% in terms of job completion time for five representative graph algorithms.