SEARCH · Engineering Papers
Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.
Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.
Asynchronous I/O is becoming increasingly popular with the large amount of data access required by scientific applications. They can take advantage of an asynchronous interface by scheduling I/O as early as possible and overlap computation or communication with I/O operations, which hides the cost associated with I/O and improves the overall performance.
DTX Explorer is a tool to generate interactive data visualizations of Darshan I/O traces collected from HPC applications. Its goal is to provide an easy and interactive way for researchers and developers to explore their application's I/O behavior and detect possible I/O bottlenecks that are impacting performance.
DTX Explorer is a tool to generate interactive data visualizations of Darshan I/O traces collected from HPC applications. Its goal is to provide an easy and interactive way for researchers and developers to explore their application's I/O behavior and detect possible I/O bottlenecks that are impacting performance.
This paper presents a twin-page storage method, which is an alternative to the TWIST (twin slot) approach by Reuter, for rapid transaction-undo recovery. In contrast to TWIST, our twin-page approach allows dirty pages in the buffer to be written at any instant onto disk without the requirement of undo logging, and, when a transaction is aborted, no explicit undo is required. As a result, all locks accumulated by the aborted transaction can be released earlier, allowing other transactions waiting for the locks to proceed. Through maintenance of aborted transaction identifiers, invalid pages written by the aborted transaction coexist with other valid pages and are guaranteed not be accessed by subsequent transactions. Instead of an explicit undo, most of the invalid pages are overwritten by subsequent normal updates. Performance in terms of disk I/O and CPU overhead for transaction-undo recovery is analyzed and compared with TWIST. It is shown that our scheme is particularly suited for applications where there are a large number of updates written onto disk when transactions are aborted, and there are frequent aborts. The approach, however, is not as applicable to environments where transactions are typically short or rarely aborted, or most updates are not written onto disk before a commitment.
High-performance parallel file systems are needed to satisfy tremendous I/O requirements of parallel scientific applications. The design of such high-performance parallel file systems depends on a comprehensive understanding of the expected workload, but so far there have been very few usage studies of multiprocessor file systems. This paper is part of the CHARISMA project, which intends to fill this void by measuring real file-system workloads on various production parallel machines. In particular, we present results from the CM-5 at the National Center for Supercomputing Applications. Our results are unique because we collect information about nearly every individual I/O request from the mix of jobs running on the machine. Analysis of the traces leads to various recommendations for parallel file-system design.
The results of single-event upset (SEU) testing performed to evaluate the parametric transients, i.e., amplitude and duration, in several I/O devices, and the impact of these transients are discussed. The failure rate of these devices is dependent on the susceptibility of interconnected devices to the resulting transient change in the output of the I/O device. This failure rate, which is a function of the susceptibility of the interconnected device as well as the SEU response of the I/O device itself, may be significantly different from an upset rate calculated without taking these factors into account. The impact at the system level is discussed by way of an example.
High performance computing (HPC) is no longer solely limited to traditional workloads such as simulation and modeling. With the increase in the popularity of machine learning (ML) and deep learning (DL) technologies, we are observing that an increasing number of HPC users are incorporating ML methods into their workflow and scientific discovery processes, across a wide spectrum of science domains such as biology, earth science, and physics. This gives rise to a diverse set of I/O patterns than the traditional checkpoint/restart-based HPC I/O behavior. The details of the I/O characteristics of such ML I/O workloads have not been studied extensively for large-scale leadership HPC systems. This paper aims to fill that gap by providing an in-depth analysis to gain an understanding of the I/O behavior of ML I/O workloads using darshan - an I/O characterization tool designed for lightweight tracing and profiling. We study the darshan logs of more than 23, 000 HPC ML I/O jobs over a time period of one year running on Summit - the second-fastest supercomputer in the world. This paper provides a systematic I/O characterization of ML I/O jobs running on a leadership scale supercomputer to understand how the I/O behavior differs across science domains and the scale of workloads, and analyze the usage of parallel file system and burst buffer by ML I/O workloads.
Monitoring and analyzing a wide range of I/O activities in an HPC cluster is important in maintaining mission-critical performance in a large-scale, multi-user, parallel storage system. Center-wide I/O traces can provide high-level information and fine-grained activities per application or per user running in the system. Studying such large-scale traces can provide helpful insights into the system. It can be used to develop predictive methods for making predictive decisions, adjusting scheduling policies, or providing decisions for the design of next-generation systems. However, sharing real-world I/O traces to expedite such research efforts leaves a few concerns; i) the cost of sharing the large traces is expensive due to this large size, and ii) privacy concern is an issue.We address such issues by building an end-to-end machine learn- ing (ML) workflow that can generate I/O traces for large-scale HPC applications. We leverage ML based feature selection and gener- ative models for I/O trace generation. The generative models are trained on I/O traces collected by the darshan I/O characterization tool over a period of one year. We present a two-step generation process consisting of two deep-learning models, called the feature generator and the trace generator. The combination of two-step generative models provides robustness by reducing the bias of the model and accounting for the stochastic nature of the I/O traces across different runs of an application. We evaluate the performance of the generative models and show that the two-step model can generate time-series I/O traces with less than 20% root mean square error.
Distributed scientific workflows are becoming more important with the interest in incorporating AI into their loops. A critical programming and performance question is how to compose workflow tasks when data is produced on one system but must be consumed on another. Since the dominant technique is composition with remote I/O, this paper explores its performance expectations. We describe BigFlowSim, a workflow I/O simulator that captures key implementation choices for remote I/O, including intensity, reuse, locality, access pattern, and data movement.With BigFlowSim, we generate a synthetic benchmark. We quantify the effects of each parameter with a performance sensitivity study. We explain trends in terms of data movement reduction and show that, under certain conditions, it is possible to establish a total order among most parameters. We apply these insights to a high energy physics workflow, Belle II Monte Carlo and simulate several I/O optimizations. Speedups range from 5% to 2×, without changing compute time.
This thesis studies the relationship by creating a tool (FTAPE) that integrates a high stress workload generator with fault injection and by using the tool to evaluate system performance under error conditions. The workloads are comprised of processes which are formed from atomic components that represent CPU, memory, and I/O activity. The fault injector is software-implemented and is capable of injecting any memory addressable location, including special registers and caches. This tool has been used to study a Tandem Integrity S2 Computer. Workloads with varying numbers of processes and varying compositions of CPU, memory, and I/O activity are first characterized in terms of performance. Then faults are injected into these workloads. The results show that as the number of concurrent processes increases, the mean fault latency initially increases due to increased contention for the CPU. However, for even higher numbers of processes (less than 3 processes), the mean latency decreases because long latency faults are paged out before they can be activated.
Large-scale simulations on nonuniform particle distributions that evolve over time are widely used in cosmology, molecular dynamics, and engineering. Such data are often saved in an unstructured format that neither preserves spatial locality nor provides metadata for accelerating spatial or attribute subset queries, leading to poor performance of visualization tasks. Furthermore, the parallel I/O strategy used typically writes a file per process or a single shared file, neither of which is portable or scalable across different HPC systems. We present a portable technique for scalable, spatially aware adaptive aggregation that preserves spatial locality in the output. We evaluate our approach on two supercomputers, Stampede2 and Summit, and demonstrate that it outperforms prior approaches at scale, achieving up to 2.5× faster writes and reads for nonuniform distributions. Furthermore, the layout written by our method is directly suitable for visual analytics, supporting low-latency reads and attribute-based filtering with little overhead.
openPMD is an open meta-data schema that provides meaning and self-description to data sets in science and engineering. The openPMD-CCD software module adds interfaces to organize camera (CCD) images in hierarchical data files. This software provides modern I/O storage formats from high-performance computing and provides bindings for integration into experimental control systems, e.g. via Python 3 and LabView 2020.
PC-CUBE is an ensemble of IBM PCs or close compatibles connected in the hypercube topology with ordinary computer cables. Communication occurs at the rate of 115.2 K-band via the RS-232 serial links. Available for PC-CUBE is the Crystalline Operating System III (CrOS III), Mercury Operating System, CUBIX and PLOTIX which are parallel I/O and graphics libraries. A CrOS performance monitor was developed to facilitate the measurement of communication and computation time of a program and their effects on performance. Also available are CXLISP, a parallel version of the XLISP interpreter; GRAFIX, some graphics routines for the EGA and CGA; and a general execution profiler for determining execution time spent by program subroutines. PC-CUBE provides a programming environment similar to all hypercube systems running CrOS III, Mercury and CUBIX. In addition, every node (personal computer) has its own graphics display monitor and storage devices. These allow data to be displayed or stored at every processor, which has much instructional value and enables easier debugging of applications. Some application programs which are taken from the book Solving Problems on Concurrent Processors (Fox 88) were implemented with graphics enhancement on PC-CUBE. The applications range from solving the Mandelbrot set, Laplace equation, wave equation, long range force interaction, to WaTor, an ecological simulation.
In the era of petascale computing, more scientific applications are being deployed on leadership scale computing platforms to enhance the scientific productivity. Many I/O techniques have been designed to address the growing I/O bottleneck on large-scale systems by handling massive scientific data in a holistic manner. While such techniques have been leveraged in a wide range of applications, they have not been shown as adequate for many mission critical applications, particularly in data post-processing stage. One of the examples is that some scientific applications generate datasets composed of a vast amount of small data elements that are organized along many spatial and temporal dimensions but require sophisticated data analytics on one or more dimensions. Including such dimensional knowledge into data organization can be beneficial to the efficiency of data post-processing, which is often missing from exiting I/O techniques. In this study, we propose a novel I/O scheme named STAR (Spatial and Temporal AggRegation) to enable high performance data queries for scientific analytics. STAR is able to dive into the massive data, identify the spatial and temporal relationships among data variables, and accordingly organize them into an optimized multi-dimensional data structure before storing to the storage. This technique not only facilitates the common access patterns of data analytics, but also further reduces the application turnaround time. In particular, STAR is able to enable efficient data queries along the time dimension, a practice common in scientific analytics but not yet supported by existing I/O techniques. In our case study with a critical climate modeling application GEOS-5, the experimental results on Jaguar supercomputer demonstrate an improvement up to 73 times for the read performance compared to the original I/O method.
The use and acceptance of new high-performance, parallel computing platforms will be impeded by the absence of an infrastructure capable of supporting orders-of-magnitude improvement in hierarchical storage and high-speed I/O (Input/Output). The distribution of these high-performance platforms and supporting infrastructures across a wide-area network further compounds this problem. We describe an architectural design and phased implementation plan for a distributed, Cooperative Storage Environment (CSE) to achieve the necessary performance, user transparency, site autonomy, communication, and security features needed to support the Accelerated Strategic Computing Initiative (ASCI). ASCI is a Department of Energy (DOE) program attempting to apply terascale platforms and Problem-Solving Environments (PSEs) toward real-world computational modeling and simulation problems. The ASCI mission must be carried out through a unified, multilaboratory effort, and will require highly secure, efficient access to vast amounts of data. The CSE provides a logically simple, geographically distributed, storage infrastructure of semi-autonomous cooperating sites to meet the strategic ASCI PSE goal of highperformance data storage and access at the user desktop.
As the input/output (I/O) needs of parallel scientific applications increase, file systems for multiprocessors are being designed to provide applications with parallel access to multiple disks. Many parallel file systems present applications with a conventional Unix-like interface that allows the application to access multiple disks transparently. This interface conceals the parallism within the file system, which increases the ease of programmability, but makes it difficult or impossible for sophisticated programmers and libraries to use knowledge about their I/O needs to exploit that parallelism. Furthermore, most current parallel file systems are optimized for a different workload than they are being asked to support. We introduce Galley, a new parallel file system that is intended to efficiently support realistic parallel workloads. Initial experiments, reported in this paper, indicate that Galley is capable of providing high-performance 1/O to applications the applications that rely on them. In Section 3 we describe that access data in patterns that have been observed to be common.
We have leveraged STARE indexing to package partitioned data chunks from diverse datasets into netCDF files, distributed them on a cluster of 16 lightweight nodes with their placements spatiotemporally co-aligned, and demonstrated a few integrative analyses using netCDF parallel I/O and Python MPI, with single-user performance and scalability comparable to, or even better than, that of a parallel array database management system (ADBMS) such as SciDB. However, records of the node location and STARE index ranges for each data chunk, similar to the chunk maps of SciDB, must be maintained and consulted by the I/O and analysis code for coordinating the analytic operations in parallel, in order to achieve the good performance and scalability.
There are now over one million UNIX sites and the pace at which new installations are added is steadily increasing. Along with this increase, comes a need to develop simple efficient, effective and adaptable ways of simultaneously collecting real-time diagnostic and performance data. This need exists because distributed systems can give rise to complex failure situations that are often un-identifiable with single-machine diagnostic software. The simultaneous collection of error and performance data is also important for research in failure prediction and error/performance studies. This paper introduces a portable method to concurrently collect real-time diagnostic and performance data on a distributed UNIX system. The combined diagnostic/performance data collection is implemented on a distributed multi-computer system using SUN4's as servers. The approach uses existing UNIX system facilities to gather system dependability information such as error and crash reports. In addition, performance data such as CPU utilization, disk usage, I/O transfer rate and network contention is also collected. In the future, the collected data will be used to identify dependability bottlenecks and to analyze the impact of failures on system performance.