Engineering Papers⌕ Search

SEARCH · Engineering Papers

Results for “Parallel file systems”

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.

At least 253 records · Page 14

Uncertainty Analysis of RELAP5-3D©

As world-wide energy consumption continues to increase, so does the demand for the use of alternative energy sources, such as Nuclear Energy. Nuclear Power Plants currently supply over 370 gigawatts of electricity, and more than 60 new nuclear reactors have been commissioned by 15 different countries. The primary concern for Nuclear Power Plant operation and lisencing has been safety. The safety of the operation of Nuclear Power Plants is no simple matter- it involves the training of operators, design of the reactor, as well as equipment and design upgrades throughout the lifetime of the reactor, etc. To safely design, operate, and understand nuclear power plants, industry and government alike have relied upon the use of best-estimate simulation codes, which allow for an accurate model of any given plant to be created with well-defined margins of safety. The most widely used of these best-estimate simulation codes in the Nuclear Power industry is RELAP5-3D. Our project focused on improving the modeling capabilities of RELAP5-3D by developing uncertainty estimates for its calculations. This work involved analyzing high, medium, and low ranked phenomena from an INL PIRT on a small break Loss-Of-Coolant Accident as wall as an analysis of a large break Loss-Of- Coolant Accident. Statistical analyses were performed using correlation coefficients. To perform the studies, computer programs were written that modify a template RELAP5 input deck to produce one deck for each combination of key input parameters. Python scripting enabled the running of the generated input files with RELAP5-3D on INL’s massively parallel cluster system. Data from the studies was collected and analyzed with SAS. A summary of the results of our studies are presented.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Office automation: The administrative window into the integrated DBMS

In parallel to the evolution of Management Information Systems from simple data files to complex data bases, the stand-alone computer systems have been migrating toward fully integrated systems serving the work force. The next major productivity gain may very well be to make these highly sophisticated working level Data Base Management Systems (DMBS) serve all levels of management with reports of varying levels of detail. Most attempts by the DBMS development organization to provide useful information to management seem to bog down in the quagmire of competing working level requirements. Most large DBMS development organizations possess three to five year backlogs. Perhaps Office Automation is the vehicle that brings to pass the Management Information System that really serves management. A good office automation system manned by a team of facilitators seeking opportunities to serve end-users could go a long way toward defining a DBMS that serves management. This paper will briefly discuss the problems of the DBMS organization, alternative approaches to solving some of the major problems, a debate about problems that may have no solution, and finally how office automation fits into the development of the Manager's Management Information System.

Brock, G. H.↗

Distributed Virtual System (DIVIRS) project

As outlined in the continuation proposal 92-ISI-50R (revised) on NASA cooperative agreement NCC 2-539, the investigators are developing software, including a system manager and a job manager, that will manage available resources and that will enable programmers to develop and execute parallel applications in terms of a virtual configuration of processors, hiding the mapping to physical nodes; developing communications routines that support the abstractions implemented; continuing the development of file and information systems based on the Virtual System Model; and incorporating appropriate security measures to allow the mechanisms developed to be used on an open network. The goal throughout the work is to provide a uniform model that can be applied to both parallel and distributed systems. The authors believe that multiprocessor systems should exist in the context of distributed systems, allowing them to be more easily shared by those that need them. The work provides the mechanisms through which nodes on multiprocessors are allocated to jobs running within the distributed system and the mechanisms through which files needed by those jobs can be located and accessed.

Schorr, Herbert↗

I/O in Machine Learning Applications on HPC Systems: A 360-degree Survey

Growing interest in Artificial Intelligence (AI) has resulted in a surge in demand for faster methods of Machine Learning (ML) model training and inference. This demand for speed has prompted the use of high performance computing (HPC) systems that excel in managing distributed workloads. Because data is the main fuel for AI applications, the performance of the storage and I/O subsystem of HPC systems is critical. In the past, HPC applications accessed large portions of data written by simulations or experiments or ingested data for visualizations or analysis tasks. ML workloads perform small reads spread across a large number of random files. This shift of I/O access patterns poses several challenges to modern parallel storage systems. In this paper, we survey I/O in ML applications on HPC systems, and target literature within a 6-year time window from 2019 to 2024. We define the scope of the survey, provide an overview of the common phases of ML, review available profilers and benchmarks, examine the I/O patterns encountered during offline data preparation, training, and inference, and explore I/O optimizations utilized in modern ML frameworks and proposed in recent literature. Lastly, we seek to expose research gaps that could spawn further R&D.

97 MATHEMATICS AND COMPUTING↗

Application-Controlled Parallel Asynchronous Input/Output Utility

A software utility tool has been designed to alleviate file system I/O performance bottlenecks to which many high-end computing (HEC) applications fall prey because of the relatively large volume of data generated for a given amount of computational work. In an effort to reduce computing resource waste, and to improve sustained performance of these HEC applications, a lightweight software utility has been designed to circumvent bandwidth limitations of typical HEC file systems by exploiting the faster inter-processor bandwidth to move output data from compute nodes to designated I/O nodes as quickly as possible, thereby minimizing the I/O wait time. This utility has successfully demonstrated a significant performance improvement within a major NASA weather application.

Clune, Thomas↗

Parallelism in System Tools

The Cray XT, when employed in conjunction with the Lustre filesystem, has provided the ability to generate huge amounts of data in the form of many files. Typically, this is accommodated by satisfying the requests of large numbers of Lustre clients in parallel. In contrast, a single service node (Lustre client) cannot adequately service such datasets. This means that the use of traditional UNIX tools like cp, tar, et alli (with have no parallel capability) can result in substantial impact to user productivity. For example, to copy a 10 TB dataset from the service node using cp would take about 24 hours, under more or less ideal conditions. During production operation, this could easily extend to 36 hours. In this paper, we introduce the Lustre User Toolkit for Cray XT, developed at the Oak Ridge Leadership Computing Facility (OLCF). We will show that Linux commands, implementing highly parallel I/O algorithms, provide orders of magnitude greater performance, greatly reducing impact to productivity.

97 MATHEMATICS AND COMPUTING↗

lustre_fssync

lustre_fssync is a tool used to sync an entire Lustre file system from one cluster to another. It uses Lustre user quotas, parallel I/O utilities and the SLURM resource manager to balance and synchronize all contents between systems.

Harr, CameronM↗

Network issues for large mass storage requirements

File Servers and Supercomputing environments need high performance networks to balance the I/O requirements seen in today's demanding computing scenarios. UltraNet is one solution which permits both high aggregate transfer rates and high task-to-task transfer rates as demonstrated in actual tests. UltraNet provides this capability as both a Server-to-Server and Server-to-Client access network giving the supercomputing center the following advantages highest performance Transport Level connections (to 40 MBytes/sec effective rates); matches the throughput of the emerging high performance disk technologies, such as RAID, parallel head transfer devices and software striping; supports standard network and file system applications using SOCKET's based application program interface such as FTP, rcp, rdump, etc.; supports access to the Network File System (NFS) and LARGE aggregate bandwidth for large NFS usage; provides access to a distributed, hierarchical data server capability using DISCOS UniTree product; supports file server solutions available from multiple vendors, including Cray, Convex, Alliant, FPS, IBM, and others.

Perdue, James↗

Distributed Virtual System (DIVIRS) Project

As outlined in our continuation proposal 92-ISI-50R (revised) on contract NCC 2-539, we are (1) developing software, including a system manager and a job manager, that will manage available resources and that will enable programmers to program parallel applications in terms of a virtual configuration of processors, hiding the mapping to physical nodes; (2) developing communications routines that support the abstractions implemented in item one; (3) continuing the development of file and information systems based on the virtual system model; and (4) incorporating appropriate security measures to allow the mechanisms developed in items 1 through 3 to be used on an open network. The goal throughout our work is to provide a uniform model that can be applied to both parallel and distributed systems. We believe that multiprocessor systems should exist in the context of distributed systems, allowing them to be more easily shared by those that need them. Our work provides the mechanisms through which nodes on multiprocessors are allocated to jobs running within the distributed system and the mechanisms through which files needed by those jobs can be located and accessed.

Schorr, Herbert↗

...And Eat it Too: High Read Performance in Write-Optimized HPC I/O Middleware File Formats

As HPC applications run on increasingly high process counts on larger and larger machines, both the frequency of checkpoints needed for fault tolerance and the resolution and size of Data Analysis Dumps are expected to increase proportionally. In order to maintain an acceptable ratio of time spent performing useful computation work to time spent performing I/O, write bandwidth to the underlying storage system must increase proportionally to this increase in the checkpoint and computation size. Unfortunately, popular scientific self-describing file formats such as netCDF and HDF5 are designed with a focus on portability and flexibility. Extra care and careful crafting of the output structure and API calls is required to optimize for write performance using these APIs. To provide sufficient write bandwidth to continue to support the demands of scientific applications, the HPC community has developed a number of I/O middleware layers, that structure output into write-optimized file formats. However, the obvious concern with any write optimized file format would be a corresponding penalty on reads. In the log-structured filesystem, for example, a file generated by random writes could be written efficiently, but reading the file back sequentially later would result in very poor performance. Simulation results require efficient read-back for visualization and analytics, and though most checkpoint files are never used, the efficiency of a restart is very important in the face of inevitable failures. The utility of write speed improving middleware would be greatly diminished if it sacrificed acceptable read performance. In this paper we examine the read performance of two write-optimized middleware layers on large parallel machines and compare it to reading data natively in popular file formats. The two I/O middleware layer examined in this paper are the Adaptable IO System (ADIOS), a library-based approach developed at Oak Ridge National Laboratory to provide a high-level IO API that can be used in place of netCDF or HDF5 to do much more aggressive write-behind and efficient reordering of data locations within the file; and the Parallel Log-structured Filesystem (PLFS), a stackable FUSE filesystem approach developed at Los Alamos National Laboratory that decouples concurrent writes to improve the speed of checkpoints. Since ADIOS is an I/O componentization that affords selection of different I/O methods at or during runtime, through a single API, users can have access to MPI-IO, Posix-IO, HDF5, netCDF, and staging methods. The ADIOS BP file format is a new log-file format that has a superset of the features of both HDF5 and netCDF, but is designed to be portable and flexible while being optimized for writing. PLFS, on the other hand, is mounted as a stackable filesystem on top of an existing parallel filesystem. Reads or writes to the PLFS filesystem are transparently translated into operations on per-process log files stored in the underlying parallel filesystem. Since PLFS performs this translation without application modification, users can write in HDF5, netCDF, or app-specific file formats and PLFS will store the writes in a set of efficiently written log-formatted files, while presenting the user with a logical 'flat' file on reads. Despite their different approaches, the commonality behind both of these middleware systems is that they both write to a log file format. As shown in previous publications, writes are fully optimized in both systems, sometimes resulting in 100x improvements over writing data in popular file formats. But as mentioned above, writing is only one part of the story. In this paper we examine the read performance of our middleware layers on large parallel machines and compare these to reading data either natively or from other popular file formats. We compare the reading performance in two different scenarios: (1) Reading back restarts from the same number of processors as wrote the data and (2) Reading back restart data from a different number of processors. We observe that not only can write-optimized I/O middleware be built to not penalize read speeds, but for important workloads, techniques that improve write performance can, perhaps counterintuitively, improve read speeds over reading from a contiguously organized file format. In the remainder of this paper, we investigate this further through case studies of PLFS and ADIOS on simulation checkpoint restarts.

99 GENERAL AND MISCELLANEOUS↗

DIstributed VIRtual System (DIVIRS) project

As outlined in our continuation proposal 92-ISI-. OR (revised) on NASA cooperative agreement NCC2-539, we are (1) developing software, including a system manager and a job manager, that will manage available resources and that will enable programmers to develop and execute parallel applications in terms of a virtual configuration of processors, hiding the mapping to physical nodes; (2) developing communications routines that support the abstractions implemented in item one; (3) continuing the development of file and information systems based on the Virtual System Model; and (4) incorporating appropriate security measures to allow the mechanisms developed in items 1 through 3 to be used on an open network. The goal throughout our work is to provide a uniform model that can be applied to both parallel and distributed systems. We believe that multiprocessor systems should exist in the context of distributed systems, allowing them to be more easily shared by those that need them. Our work provides the mechanisms through which nodes on multiprocessors are allocated to jobs running within the distributed system and the mechanisms through which files needed by those jobs can be located and accessed.

Schorr, Herbert↗

DIstributed VIRtual System (DIVIRS) project

As outlined in our continuation proposal 92-ISI-50R (revised) on NASA cooperative agreement NCC2-539, we are (1) developing software, including a system manager and a job manager, that will manage available resources and that will enable programmers to develop and execute parallel applications in terms of a virtual configuration of processors, hiding the mapping to physical nodes; (2) developing communications routines that support the abstractions implemented in item one; (3) continuing the development of file and information systems based on the Virtual System Model; and (4) incorporating appropriate security measures to allow the mechanisms developed in items 1 through 3 to be used on an open network. The goal throughout our work is to provide a uniform model that can be applied to both parallel and distributed systems. We believe that multiprocessor systems should exist in the context of distributed systems, allowing them to be more easily shared by those that need them. Our work provides the mechanisms through which nodes on multiprocessors are allocated to jobs running within the distributed system and the mechanisms through which files needed by those jobs can be located and accessed.

Schorr, Herbert↗

Reverse Computation for Rollback-based Fault Tolerance in Large Parallel Systems

Reverse computation is presented here as an important future direction in addressing the challenge of fault tolerant execution on very large cluster platforms for parallel computing. As the scale of parallel jobs increases, traditional checkpointing approaches suffer scalability problems ranging from computational slowdowns to high congestion at the persistent stores for checkpoints. Reverse computation can overcome such problems and is also better suited for parallel computing on newer architectures with smaller, cheaper or energy-efficient memories and file systems. Initial evidence for the feasibility of reverse computation in large systems is presented with detailed performance data from a particle simulation scaling to 65,536 processor cores and 950 accelerators (GPUs). Reverse computation is observed to deliver very large gains relative to checkpointing schemes when nodes rely on their host processors/memory to tolerate faults at their accelerators. A comparison between reverse computation and checkpointing with measurements such as cache miss ratios, TLB misses and memory usage indicates that reverse computation is hard to ignore as a future alternative to be pursued in emerging architectures.

Perumalla, Kalyan S↗

DSN Beowulf Cluster-Based VLBI Correlator

The NASA Deep Space Network (DSN) requires a broadband VLBI (very long baseline interferometry) correlator to process data routinely taken as part of the VLBI source Catalogue Maintenance and Enhancement task (CAT M&E) and the Time and Earth Motion Precision Observations task (TEMPO). The data provided by these measurements are a crucial ingredient in the formation of precision deep-space navigation models. In addition, a VLBI correlator is needed to provide support for other VLBI related activities for both internal and external customers. The JPL VLBI Correlator (JVC) was designed, developed, and delivered to the DSN as a successor to the legacy Block II Correlator. The JVC is a full-capability VLBI correlator that uses software processes running on multiple computers to cross-correlate two-antenna broadband noise data. Components of this new system (see Figure 1) consist of Linux PCs integrated into a Beowulf Cluster, an existing Mark5 data storage system, a RAID array, an existing software correlator package (SoftC) originally developed for Delta DOR Navigation processing, and various custom- developed software processes and scripts. Parallel processing on the JVC is achieved by assigning slave nodes of the Beowulf cluster to process separate scans in parallel until all scans have been processed. Due to the single stream sequential playback of the Mark5 data, some ramp-up time is required before all nodes can have access to required scan data. Core functions of each processing step are accomplished using optimized C programs. The coordination and execution of these programs across the cluster is accomplished using Pearl scripts, PostgreSQL commands, and a handful of miscellaneous system utilities. Mark5 data modules are loaded on Mark5 Data systems playback units, one per station. Data processing is started when the operator scans the Mark5 systems and runs a script that reads various configuration files and then creates an experiment-dependent status database used to delegate parallel tasks between nodes and storage areas (see Figure 2). This script forks into three processes: extract, translate, and correlate. Each of these processes iterates on available scan data and updates the status database as the work for each scan is completed. The extract process coordinates and monitors the transfer of data from each of the Mark5s to the Beowulf RAID storage systems. The translate process monitors and executes the data conversion processes on available scan files, and writes the translated files to the slave nodes. The correlate process monitors the execution of SoftC correlation processes on the slave nodes for scans that have completed translation. A comparison of the JVC and the legacy Block II correlator outputs reveals they are well within a formal error, and that the data are comparable with respect to their use in flight navigation. The processing speed of the JVC is improved over the Block II correlator by a factor of 4, largely due to the elimination of the reel-to-reel tape drives used in the Block II correlator.

Rogstad, Stephen P.↗

Template based parallel checkpointing in a massively parallel computer system

A method and apparatus for a template based parallel checkpoint save for a massively parallel super computer system using a parallel variation of the rsync protocol, and network broadcast. In preferred embodiments, the checkpoint data for each node is compared to a template checkpoint file that resides in the storage and that was previously produced. Embodiments herein greatly decrease the amount of data that must be transmitted and stored for faster checkpointing and increased efficiency of the computer system. Embodiments are directed to a parallel computer system with nodes arranged in a cluster with a high speed interconnect that can perform broadcast communication. The checkpoint contains a set of actual small data blocks with their corresponding checksums from all nodes in the system. The data blocks may be compressed using conventional non-lossy data compression algorithms to further reduce the overall checkpoint size.

97 MATHEMATICS AND COMPUTING↗

Distributed File System Utilities to Manage Large DatasetsVersion 0.5

FileUtils provides a suite of tools to manage large datasets typically created by large parallel MPI applications. They are written in C and use standard POSIX I/Ocalls. The current suite consists of tools to copy, compare, remove, and list. The tools provide dramatic speedup over existing Linux tools, which often run as a single process.

ECP↗

Initial Characterization of Parallel NFS Implementations

Parallel NFS (pNFS) is touted as an emergent standard protocol for parallel I/O access in various storage environments. Several pNFS prototypes have been implemented for initial validation and protocol examination. Previous efforts have focused on realizing the pNFS protocol to expose the best bandwidth potential from underlying file and storage systems. In this presentation, we provide an initial characterization of two pNFS prototype implementations, lpNFS (a Lustre-based parallel NFS implementation) and spNFS (another reference implementation from Network Appliance, Inc.). We show that both lpNFS and spNFS can faithfully achieve the primary goal of pNFS, i.e., aggregating I/O bandwidth from many storage servers. However, they both face the challenge of scalable metadata management. Particularly, the throughput of sp-NFS metadata operations degrades significanlty with an increasing number of data servers. Even for the better-performing lpNFS, we discuss its architecture and propose a direct I/O request flow protocol to improve its performance.

97 MATHEMATICS AND COMPUTING↗