Search NASA⌕ Search

Engineering topics

Xie, Bing

Publications and source records attributed to Xie, Bing.

At least 19 records

Access Patterns and Performance Behaviors of Multi-layer Supercomputer I/O Subsystems under Production Load

Scientific computing workloads at HPC facilities have been shifting from traditional numerical simulations to AI/ML applications for training and inference while processing and producing ever-increasing amounts of scientific data. To address the growing need for increased storage capacity, lower access latency, and higher bandwidth, emerging technologies such as non-volatile memory are integrated into supercomputer I/O subsystems. With these emerging trends, we need a better understanding of the multilayer supercomputer I/O systems and ways to use these subsystems efficiently. In this work, we study the I/O access patterns and performance characteristics of two representative supercomputer I/O subsystems. Through an extensive analysis of year-long I/O logs on each system, we report new observations in I/O reads and writes, unbalanced use of storage system layers, and new trends in user behaviors at the HPC I/O middleware stack.

Bez, JL↗

SchedInspector: A Batch Job Scheduling Inspector Using Reinforcement Learning

Improving the performance of job executions is an important goal of HPC batch job schedulers, such as minimizing job waiting time, slowdown, or completion time. Such a goal is often accomplished using carefully designed heuristics based on job features, such as job size and job duration. However, these heuristics overlook important runtime factors (e.g., cluster availability and waiting job patterns), which may vary across time and make a previously sound scheduling decision not hold any longer. In this study, we propose a new approach to incorporate runtime factors into batch job scheduling for better job execution performance. The key idea is to add a scheduling inspector on top of the base job scheduler to scrutinize its scheduling decisions. The inspector will take the runtime factors into consideration and accordingly determine the fitness of the scheduled job. It then either accepts the scheduled job or rejects it and asks the base schedulers to try again later. We realize such an inspector, namely SchedInspector, by leveraging the intelligence of reinforcement learning. Through extensive experiments, we show SchedInspector can intelligently integrate the runtime factors into various batch job scheduling policies, including the state-of-the-art one, to gain better job execution performance, such as smaller average bounded job slowdown (up to 69% better) or average job waiting time (up to 52% better), across various real-world workloads. We also show that although rejecting scheduling decisions may leave the resources idle hence affect the system utilization, SchedInspector is able to achieve the job execution performance improvement with marginal impact on the system utilization (typically less than 1%). We consider one key advantage of SchedInspector is it automatically learns to work with and improve existing job scheduling policies without changing them, which makes it promising to serve as a generic enhancer for various batch job scheduling policies.

Zhang, Di↗

April 2020 Darshan counters from the Summit supercomputer

This dataset is the Darshan counters collected from the Summit supercomputer in a month of April 2020. 1. Description of methods used for collection/generation of data: Job submitted on Summit HPC system when completed successfully and has made I/O calls (captured by Darshan tool) writes a Darshan log file on alpine filesystem. One job can have multiple `jsrun` commands and Darshan will generate separate logs each log corresponding to an `jsrun` command, so a job can have one or more Darshan logs associated with it. 2. Methods for processing the data: To process the data, we first use `darshan-util` tool to parse the Darshan logs. Then we restructure the logs and merge data from multiple Darshan logs if they belong to the same Summit job.

97 MATHEMATICS AND COMPUTING↗

Accelerating Collective Communication in Data Parallel Training across Deep Learning Frameworks

This work develops new techniques within Horovod, a generic communication library supporting data parallel training across deep learning frameworks. In particular, we improve the Horovod control plane by implementing a new coordination scheme that takes advantage of the characteristics of the typical data parallel training paradigm, namely the repeated execution of collectives on the gradients of a fixed set of tensors. Using a caching strategy, we execute Horovod’s existing coordinator-worker logic only once during a typical training run, replacing it with a more efficient decentralized orchestration strategy using the cached data and a global intersection of a bitvector for the remaining training duration. Next, we introduce a feature for end users to explicitly group collective operations, enabling finer grained control over the communication buffer sizes. To evaluate our proposed strategies, we conduct experiments on a world-class supercomputer — Summit. We compare our proposals to Horovod’s original design and observe 2x performance improvement at a scale of 6000 GPUs; we also compare them against tf.distribute and torch.DDP and achieve 12% better and comparable performance, respectively, using up to 1536 GPUs; we compare our solution against BytePS in typical HPC settings and achieve about 20% better performance on a scale of 768 GPUs. Finally, we test our strategies on a scientific application (STEMDL) using up to 27,600 GPUs (the entire Summit) and show that we achieve a near-linear scaling of 0.93 with a sustained performance of 1.54 exaflops (with standard error +- 0.02) in FP16 precision.

Romero, Joshua↗

Understanding and Leveraging the I/O Patterns of Emerging Machine Learning Analytics

The scientific community is currently experiencing unprecedented amounts of data generated by cutting-edge science facilities. Soon facilities will be producing up to 1 PB/s which will force scientist to use more autonomous techniques to learn from the data. The adoption of machine learning methods, like deep learning techniques, in large-scale workflows comes with a shift in the workflow’s computational and I/O patterns. These changes often include iterative processes and model architecture searches, in which datasets are analyzed multiple times in different formats with different model configurations in order to find accurate, reliable and efficient learning models. This shift in behavior brings changes in I/O patterns at the application level as well at the system level. These changes also bring new challenges for the HPC I/O teams, since these patterns contain more complex I/O workloads. In this paper we discuss the I/O patterns experienced by emerging analytical codes that rely on machine learning algorithms and highlight the challenges in designing efficient I/O transfers for such workflows. We comment on how to leverage the data access patterns in order to fetch in a more efficient way the required input data in the format and order given by the needs of the application and how to optimize the data path between collaborative processes. We will motivate our work and show performance gains with a study case of medical applications.

Gainaru, Ana↗

I/O Bottleneck Detection and Tuning: Connecting the Dots using Interactive Log Analysis

Using parallel file systems efficiently is a tricky problem due to inter-dependencies among multiple layers of I/O software, including high-level I/O libraries (HDF5, netCDF, etc.), MPI-IO, POSIX, and file systems (GPFS, Lustre, etc.). Profiling tools such as Darshan collect traces to help understand the I/O performance behavior. However, there are significant gaps in analyzing the collected traces and then applying tuning options offered by various layers of I/O software. Seeking to connect the dots between I/O bottleneck detection and tuning, we propose DXT Explorer, an interactive log analysis tool. In this paper, we present a case study using our interactive log analysis tool to identify and apply various I/O optimizations. We report an evaluation of performance improvement achieved for four I/O kernels extracted from science applications.

Bez, Jean Luca↗

WIRE: Resource-efficient Scaling with Online Prediction for DAG-based Workflows

This paper introduces WIRE that manages resources for the DAG-based workflows on IaaS clouds. WIRE predicts and plans resources over the MAPE (Monitor-Analyze-Plan-Execute) loops to: 1) Estimate task performance with online data, 2) Conduct simulations to predict the upcoming loads based on online estimates and workflow DAGs, 3) Apply a resource-steering policy to size cloud instance pools for the maximal parallelism that is consistent with low cost. We implement WIRE on Pegasus WMS/HTCondor and evaluate its performance on the ExoGENI network cloud. The results show that WIRE attains low resource cost with the performance that is typically within a factor of two of optimal.

Xie, Bing↗

Battle of the Defaults: Extracting Performance Characteristics of HDF5 under Production Load

Popular parallel I/O libraries, such as HDF5, provide tuning parameters to obtain superior performance. However, the selection of effective parameters on production systems is complex due to the interdependence of I/O software and file system layers. Hence, application developers typically use the default parameters and often experience poor I/O performance. This work conducts a benchmarking-based analysis on the HDF5 behaviors with a wide variety of I/O patterns to extract performance characteristics under the production workload. To make the analysis well controlled, we exercise I/O benchmarks on POSIX-IO, MPI-IO, and HDF5 using the same I/O patterns and in the same jobs. To address high performance variability in production environments, we repeat the benchmarks across I/O patterns, storage devices, and time intervals. Based on the results, we identified consistent HDF5 behaviors that appropriate configurations and operations on dataset layout and file-metadata placement can improve performance significantly. We apply our findings and evaluate the tuned I/O library on two supercomputers: Summit and Cori. The results show that our tuned parameters can achieve more than 10× I/O performance speedup than that with default parameters on both systems, suggesting the effectiveness, stability, and generality of our solution.

Xie, Bing↗

Interpreting Write Performance of Supercomputer I/O Systems with Regression Models

This work seeks to advance the state of the art in HPC I/O performance analysis and interpretation. In particular, we demonstrate effective techniques to: (1) model output performance in the presence of I/O interference from production loads; (2) build features from write patterns and key parameters of the system architecture and configurations; (3) employ suitable machine learning algorithms to improve model accuracy. We train models with five popular regression algorithms and conduct experiments on two distinct production HPC platforms. We find that the lasso and random forest models predict output performance with high accuracy on both of the target systems. We also explore use of the models to guide adaptation in I/O middleware systems, and show potential for improvements of at least 15% from model-guided adaptation on 70% of samples, and improvements up to 10× on some samples for both of the target systems.

Xie, Bing↗

RLScheduler: An Automated HPC Batch Job Scheduler Using Reinforcement Learning

Today’s high-performance computing (HPC) platforms are still dominated by batch jobs. Accordingly, effective batch job scheduling is crucial to obtain high system efficiency. Existing HPC batch job schedulers typically leverage heuristic priority functions to prioritize and schedule jobs. But, once configured and deployed by the experts, such priority functions can hardly adapt to the changes of job loads, optimization goals, or system settings, potentially leading to degraded system efficiency when changes occur. To address this fundamental issue, we present RLScheduler, an automated HPC batch job scheduler built on reinforcement learning. RLScheduler relies on minimal manual interventions or expert knowledge, but can learn high-quality scheduling policies via its own continuous ‘trial and error’. We introduce a new kernel-based neural network structure and trajectory filtering mechanism in RLScheduler to improve and stabilize the learning process. Through extensive evaluations, we confirm that RLScheduler can learn high-quality scheduling policies towards various workloads and various optimization goals with relatively low computation cost. Moreover, we show that the learned models perform stably even when applied to unseen workloads, making them practical for production use.

Zhang, Di↗

US Department of Energy, Office of Science High Performance Computing Facility Operational Assessment 2019 Oak Ridge Leadership Computing Facility

Oak Ridge National Laboratory's (ORNL's) Leadership Computing Facility (OLCF) continues to surpass its operational target goals: supporting users; delivering fast, reliable computational ecosystems; creating innovative solutions for high performance computing (HPC) needs; and managing risks, safety, and security associated with operating some of the most powerful computers in the world. The results can be seen in the cutting-edge science conducted by users and the praise from the research community. Calendar year (CY) 2019 was a big year as OLCF staff ran five world-class resources (the leadershipclass computers Titan and Summit, the large analysis cluster called Eos, and the massive parallel filesystems called Atlas and Alpine)) and also began power and cooling upgrades for a 2021 exascale system called Frontier. While continuing exceptional operation of Titan, Eos, and Rhea, the OLCF released the Summit supercomputer for production on January 1, 2019. Summit debuted as the most capable and efficient system in its class and has been recognized as the most powerful system in the world for its performance on both the high performance linpack (HPL) and conjugate gradient (HPCG) benchmark applications since June 2018 according to TOP500. Summit represents the culmination of a multiyear effort between the OLCF, IBM, NVIDIA, and Mellanox to deliver a system that is unmatched for modeling, simulation, data analysis, and learning. To hit the ground running with science-ready applications on day one, application teams worked closely with the OLCF through the Center for Accelerated Application Readiness (CAAR) program for years in advance of the Summit deployment. CY 2019 was filled with outstanding results and accomplishments: a very high rating from users on overall satisfaction for the sixth year in a row; a tremendous amount of core-hours delivered to researchers from two leadership-class systems; and success in delivering on the allocation split of roughly 60%, 30%, and 10% of core-hours offered for the Innovative and Novel Computational Impact on Theory and Experiment (INCITE), Advanced Scientific Computing Research Leadership Computing Challenge (ALCC), and Director's Discretionary (DD) programs, respectively (see Operational Performance section). These accomplishments, coupled with the high utilization rates (overall and capability usage), represent the fulfillment of the promise of both leadership-class machines: efficient facilitation of leadership-class computational applications. Table ES.1 presents a summary of the 2019 OLCF metric targets and the associated results. More information can be found in the Operational Performance section for each OLCF resource. The scientific accomplishments of OLCF users are a strong indication of long-term operational success, with publications this year in such notable journals and publications as Nature, Nature Physics, Nature Plants, Physical Review X, Journal of the American Physical Society, Cell, Nano Letters, and Trends in Biotechnology. Crucial domain-specific discoveries facilitated by resources at the OLCF are described in the High Performance Computing Facility Operational Assessment 2019 Oak Ridge Leadership Computing Facility (OAR) Strategic Results section. For example, researchers used Summit to pinpoint and understand the production of proteins from genetic information, including mutations and the functional expression of disease (Section 8.2).

97 MATHEMATICS AND COMPUTING↗

Characterizing Output Bottlenecks of a Production Supercomputer: Analysis and Implications

This article studies the I/O write behaviors of the Titan supercomputer and its Lustre parallel file stores under production load. The results can inform the design, deployment, and configuration of file systems along with the design of I/O software in the application, operating system, and adaptive I/O libraries.We propose a statistical benchmarking methodology to measure write performance across I/O configurations, hardware settings, and system conditions. Moreover, we introduce two relative measures to quantify the write-performance behaviors of hardware components under production load. In addition to designing experiments and benchmarking on Titan, we verify the experimental results on one real application and one real application I/O kernel, XGC and HACC IO, respectively. These two are representative and widely used to address the typical I/O behaviors of applications.In summary, we find that Titan’s I/O system is variable across the machine at fine time scales. This variability has two major implications. First, stragglers lessen the benefit of coupled I/O parallelism (striping). Peak median output bandwidths are obtained with parallel writes to many independent files, with no striping or write sharing of files across clients (compute nodes). I/O parallelism is most effective when the application—or its I/O libraries—distributes the I/O load so that each target stores files for multiple clients and each client writes files on multiple targets in a balanced way with minimal contention. Second, our results suggest that the potential benefit of dynamic adaptation is limited. In particular, it is not fruitful to attempt to identify “good locations” in the machine or in the file system: component performance is driven by transient load conditions and past performance is not a useful predictor of future performance. For example, we do not observe diurnal load patterns that are predictable.

97 MATHEMATICS AND COMPUTING↗

Applying Machine Learning to Understand Write Performance of Large-scale Parallel Filesystems

In high-performance computing (HPC), I/O performance prediction offers the potential to improve the efficiency of scientific computing. In particular, accurate prediction can make runtime estimates more precise, guide users toward optimal checkpoint strategies, and better inform facility provisioning and scheduling policies. HPC I/O performance is notoriously difficult to predict and model, however, in large part because of inherent variability and a lack of transparency in the behaviors of constituent storage system components. In this work we seek to advance the state of the art in HPC I/O performance prediction by (1) modeling the mean performance to address high variability, (2) deriving model features from write patterns, system architecture and system configurations, and (3) employing Lasso regression model to improve model accuracy. We demonstrate the efficacy of our approach by applying it to a crucial subset of common HPC I/O motifs, namely, file-per-process checkpoint write workloads. We conduct experiments on two distinct production HPC platforms—Titan at the Oak Ridge Leadership Computing Facility and Cetus at the Argonne Leadership Computing Facility—to train and evaluate our models. We find that we can attain ≤ 30% relative error for 92.79% and 99.64% of the samples in our test set on these platforms, respectively.

Xie, Bing↗

Data Jockey: Automatic Data Management for HPC Multi-Tiered Storage Systems

We present the design and implementation of Data Jockey, a data management system for HPC multi-tiered storage systems. As a centralized data management control plane, Data Jockey automates bulk data movement and placement for scientific workflows and integrates into existing HPC storage infrastructures. Data Jockey simplifies data management by eliminating human effort in programming complex data movements, laying datasets across multiple storage tiers when supporting complex workflows, which in turn increases the usability of multitiered storage systems emerging in modern HPC data centers.Specifically, Data Jockey presents a new data management scheme called “goal driven data management” that can automatically infer low-level bulk data movement plans from declarative high-level goal statements that come from the lifetime of iterative runs of scientific workflows. While doing so, Data Jockey aims to minimize data wait times by taking responsibility for datasets that are unused or to be used, and aggressively utilizing the capacity of the upper, higher performant storage tiers.We evaluated a prototype implementation of Data Jockey under a synthetic workload based on a year’s worth of Oak Ridge Leadership Computing Facility’s (OLCF) operational logs. Our evaluations suggest that Data Jockey leads to higher utilization of the upper storage tiers while minimizing the programming effort of data movement compared to human involved, per-domain adhoc data management scripts.

Shin, Woong↗

Engineering Single-Atom Cobalt Catalysts toward Improved Electrocatalysis

In this paper, the development of cost-effective catalysts to replace noble metal is attracting increasing interests in many fields of catalysis and energy, and intensive efforts are focused on the integration of transition-metal sites in carbon as noble-metal-free candidates. Recently, the discovery of single-atom dispersed catalyst (SAC) provides a new frontier in heterogeneous catalysis. However, the electrocatalytic application of SAC is still subject to several theoretical and experimental limitations. Further advances depend on a better design of SAC through optimizing its interaction with adsorbates during catalysis. Here, distinctive from previous studies, favorable 3d electronic occupation and enhanced metal-adsorbates interactions in single-atom centers via the construction of nonplanar coordination is achieved, which is confirmed by advanced X-ray spectroscopic and electrochemical studies. The as-designed atomically dispersed cobalt sites within nonplanar coordination show significantly improved catalytic activity and selectivity toward the oxygen reduction reaction, approaching the benchmark Pt-based catalysts. More importantly, the illustration of the active sites in SAC indicates metal-natured catalytic sites and a media-dependent catalytic pathway. Achieving structural and electronic engineering on SAC that promotes its catalytic performances provides a paradigm to bridge the gap between single-atom catalysts design and electrocatalytic applications.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗