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 55 records · Page 3

Rethinking key–value store for parallel I/O optimization

Key-value stores are being widely used as the storage system for large-scale internet services and cloud storage systems. However, they are rarely used in HPC systems, where parallel file systems are the dominant storage solution. In this study, we examine the architecture differences and performance characteristics of parallel file systems and key-value stores. We propose using key-value stores to optimize overall Input/Output (I/O) performance, especially for workloads that parallel file systems cannot handle well, such as the cases with intense data synchronization or heavy metadata operations. We conducted experiments with several synthetic benchmarks, an I/O benchmark, and a real application. We modeled the performance of these two systems using collected data from our experiments, and we provide a predictive method to identify which system offers better I/O performance given a specific workload. The results show that we can optimize the I/O performance in HPC systems by utilizing key-value stores.

Hyperdex↗

Small File Aggregation with PLFS [Slides]

Today’s computational science demands have resulted in ever larger parallel computers, and storage systems have grown to match these demands. Parallel file systems used in this environment are increasingly specialized to extract the highest possible performance for large I/O operations, at the expense of other potential workloads. While some applications have adapted to I/O best practices and can obtain good performance on these systems, the natural I/O patterns of many applications result in the generation of a huge number of small files, the creation of which is poorly served by current parallel file systems at very large scale. This paper describes a new technique for optimizing small file access in parallel file systems for these very large scale systems. The idea is to use a virtual parallel log-structure file system on the compute nodes in order to aggregate large numbers of small files in compute node memory and then stream their data sequentially to a much smaller number of physical files on an underlying parallel file system. The technique is implemented and evaluated using PLFS as the aggregating middleware. We evaluate our system with micro-benchmarks on a local OSX filesystem and with an MPI extension of the standard Postmark to provide results at scale on both Lustre and PanFS parallel filesystems. We observe as much as a 33x improvement in small file create rates on a single host, and 30x improvement in small file write rates, compared to a baseline Lustre configuration on a leadership computing platform using 16,384 cores and achieve an unprecedented create rate of 200 million files per second.

97 MATHEMATICS AND COMPUTING↗

Lessons Learned in Deploying the World s Largest Scale Lustre File System

The Spider system at the Oak Ridge National Laboratory's Leadership Computing Facility (OLCF) is the world's largest scale Lustre parallel file system. Envisioned as a shared parallel file system capable of delivering both the bandwidth and capacity requirements of the OLCF's diverse computational environment, the project had a number of ambitious goals. To support the workloads of the OLCF's diverse computational platforms, the aggregate performance and storage capacity of Spider exceed that of our previously deployed systems by a factor of 6x - 240 GB/sec, and 17x - 10 Petabytes, respectively. Furthermore, Spider supports over 26,000 clients concurrently accessing the file system, which exceeds our previously deployed systems by nearly 4x. In addition to these scalability challenges, moving to a center-wide shared file system required dramatically improved resiliency and fault-tolerance mechanisms. This paper details our efforts in designing, deploying, and operating Spider. Through a phased approach of research and development, prototyping, deployment, and transition to operations, this work has resulted in a number of insights into large-scale parallel file system architectures, from both the design and the operational perspectives. We present in this paper our solutions to issues such as network congestion, performance baselining and evaluation, file system journaling overheads, and high availability in a system with tens of thousands of components. We also discuss areas of continued challenges, such as stressed metadata performance and the need for file system quality of service alongside with our efforts to address them. Finally, operational aspects of managing a system of this scale are discussed along with real-world data and observations.

99 GENERAL AND MISCELLANEOUS↗

Leveraging Pre-Built Catalogs and Object-Level Scheduling to Eliminate I/O Bottlenecks in HPC Environments

Modern High-Performance Computing (HPC) environments face mounting challenges due to the shift from large to small file datasets, along with an increasing number of users and parallelized applications. As HPC systems rely on Parallel File Systems (PFS), such as Lustre for data processing, performance bottlenecks stemming from Object Storage Target (OST) contention have become a significant concern. Existing solutions, such as LADS with its object-level scheduling approach, fall short in large-scale HPC environments due to their inability to effectively address metadata I/O bottlenecks and the growing number of I/O processes. This study highlights the pressing need for a comprehensive solution that tackles both OST contention and metadata I/O challenges in diverse HPC workloads. To address these challenges, we propose SwiftLoad, an object-level I/O scheduling framework that leverages a metadata catalog to enhance the performance and efficiency of parallel HPC utilities. The adoption of the metadata catalog mitigates the metadata I/O bottlenecks that commonly occur in HPC utilities, a challenge that is particularly pronounced in object-level I/O scheduling. SwiftLoad addresses OST contention and the uneven distribution of I/O processes across different OSTs through mathematical modeling and incorporates a Loader Configuration Module to regulate the number of I/O processes. Evaluated with two representative utilities—data deduplication profiling and data augmentation—SwiftLoad achieved performance improvements of up to 5.63x and 11.0x, respectively, on a production supercomputer.

HPC↗

Diving into Petascale Production File Systems through Large Scale Profiling and Analysis

As leadership computing facilities grow their storage capacity into the multi- petabyte range, the number of files and directories leap into the scale of billions. A complete profiling of such a parallel file system in a production environment presents a unique challenge. On one hand, the time, resources, and negative performance impact on production users can make regular profiling difficult. On the other hand, the result of such profiling can yield much needed understanding of the file system's general characteristics, as well as provide insight to how users write and access their data on a grand scale. This paper presents a lightweight and scalable profiling solution that can efficiently walk, analyze, and profile multi-petabyte parallel file systems. This tool has been deployed and is in regular use on very large-scale production parallel file systems at both Oak Ridge National Lab's Oak Ridge Leadership Facility (OLCF) and Lawrence Livermore National Lab's Livermore Computing (LC) facilities. We present the results of our initial analysis on the data collected from these two large-scale production systems, organized into three use cases: (1) file system snapshot and composition, (2) striping pattern analysis for Lustre, and (3) simulated storage capacity utilization in preparation for future file systems. Our analysis shows that on the OLCF file system, over 96% of user files exhibit the default stripe width, potentially limiting performance on large files by underutilizing storage servers and disks. Our simulated block analysis quantitatively shows the space overhead when doing a forklift system migration. It also reveals that due to the difference in system compositions (OLCF vs. LC), we can achieve better performance and space trade-offs by employing different native file system block sizes.

Wang, Feiyi↗

IOrchestrator: improving the performance of multi-node I/O systems via inter-server coordination

A cluster of I/O nodes and a parallel file system are often used to provide high-throughput I/O service to a parallel compute cluster. To exploit I/O parallelism parallel file systems stripe file data across the I/O nodes. While this practice is effective in serving asynchronous requests, it may break individual program's spatial locality, which can seriously degrade I/O performance when the I/O nodes concurrently serve synchronous requests from multiple I/O-intensive programs. In this paper we propose a scheme, Orchestrator, to improve the I/O performance of multi-I/O-node systems by orchestrating I/O services among programs when such inter-I/O-node coordination is dynamically determined to be cost effective. We have implemented IOrchestrator in the PVFS2 parallel file system. Our experiments with representative parallel benchmarks show that IOrchestrator can significantly improve I/O performance - by up to a factor of 2.5 - delivered by a cluster of I/O nodes servicing concurrently-running parallel programs. Notably, we have not observed any scenarios in which the use of IOrchestrator causes significant performance degradation.

97 MATHEMATICS AND COMPUTING↗

Detailed Modeling, Design, and Evaluation of a Scalable Multi-level Checkpointing System

High-performance computing (HPC) systems are growing more powerful by utilizing more hardware components. As the system mean-time-before-failure correspondingly drops, applications must checkpoint more frequently to make progress. However, as the system memory sizes grow faster than the bandwidth to the parallel file system, the cost of checkpointing begins to dominate application run times. A potential solution to this problem is to use multi-level checkpointing, which employs multiple types of checkpoints with different costs and different levels of resiliency in a single run. The goal is to design light-weight checkpoints to handle the most common failure modes and rely on more expensive checkpoints for less common, but more severe failures. While this approach is theoretically promising, it has not been fully evaluated in a large-scale, production system context. To this end we have designed a system, called the Scalable Checkpoint/Restart (SCR) library, that writes checkpoints to storage on the compute nodes utilizing RAM, Flash, or disk, in addition to the parallel file system. We present the performance and reliability properties of SCR as well as a probabilistic Markov model that predicts its performance on current and future systems. We show that multi-level checkpointing improves efficiency on existing large-scale systems and that this benefit increases as the system size grows. In particular, we developed low-cost checkpoint schemes that are 100x-1000x faster than the parallel file system and effective against 85% of our system failures. This leads to a gain in machine efficiency of up to 35%, and it reduces the the load on the parallel file system by a factor of two on current and future systems.

99 GENERAL AND MISCELLANEOUS↗

Detailed Modeling and Evaluation of a Scalable Multilevel Checkpointing System

High-performance computing (HPC) systems are growing more powerful by utilizing more components. As the system mean time before failure correspondingly drops, applications must checkpoint frequently to make progress. But, at scale, the cost of checkpointing becomes prohibitive. A solution to this problem is multilevel checkpointing, which employs multiple types of checkpoints in a single run. Moreover, lightweight checkpoints can handle the most common failure modes, while more expensive checkpoints can handle severe failures. We designed a multilevel checkpointing library, the Scalable Checkpoint/Restart (SCR) library, that writes lightweight checkpoints to node-local storage in addition to the parallel file system. We present probabilistic Markov models of SCR's performance. We show that on future large-scale systems, SCR can lead to a gain in machine efficiency of up to 35 percent, and reduce the load on the parallel file system by a factor of two. In addition, we predict that checkpoint scavenging, or only writing checkpoints to the parallel file system on application termination, can reduce the load on the parallel file system by 20 × on today's systems and still maintain high application efficiency.

97 MATHEMATICS AND COMPUTING↗

Multi-Node Program Fuzzing on High Performance Computing Resources

Significant effort is placed on tuning the internal parameters of fuzzers to explore the state space, measured as coverage, of binaries. In this work, we investigate the effects of the external environment on the resulting coverage after fuzzing two binaries with AFL for 24 hours. Parameters such as scaling to multiple nodes, node saturation, and parallel file system type on HPC resources are controlled in order to maximize coverage. It will be shown that employing a parallel file system such as IBM's General Parallel File System offers an advantage for fuzzing operations, since it contains enhancements for performance optimization. When combined with scaling to two and four nodes, while simultaneously restricting the number of coordinated AFL tasks per node on the low end (10-50% of available physical cores), coverage may be enhanced within a shorter period of time. Thus, controlling the external environment is a useful effort.

97 MATHEMATICS AND COMPUTING↗

SLoG: Large-Scale Logging Middleware for HPC and Big Data Convergence

Cloud developers traditionally rely on purpose-specific services to provide the storage model they need for an application. In contrast, HPC developers have a much more limited choice, typically restricted to a centralized parallel file system for persistent storage. Unfortunately, these systems often offer low performance when subject to highly concurrent, conflicting I/O patterns. This makes difficult the implementation of inherently concurrent data structures such as distributed shared logs. Yet, this data structure is key to applications such as computational steering, data collection from physical sensor grids, or discrete event generators. In this paper we tackle this issue. We present SLoG, shared log middleware providing a shared log abstraction over a parallel file system, designed to circumvent the aforementioned limitations. We evaluate SLoG’s design on up to 100,000 cores of the Theta supercomputer: the results show high append velocity at scale while also providing substantial benefits for other persistent backend storage systems.

big data↗

Management and Storage of Scientific Data

Scientific discoveries rely heavily on efficient access, search, and management of massive data sets. Data management technologies have, for decades, provided foundational capabilities for scientific computing. Just as storage, input/output (I/O), and data management have been fundamental to simulation-based science for many years, so too are capable data-management technologies key to the success of today’s scientific workflows utilizing data intensive and machine learning (ML) techniques. The Department of Energy, Office of Science, Advanced Scientific Computing Research (ASCR) program has invested broadly in data-management research focused on high-performance computing (HPC) systems, from parallel file systems that store data to application software that makes these systems more productive. Still, advances in technology combined with growing diversity of supported science strongly motivate continued investment in this area. In January 2022, ASCR convened a workshop to identify priority research directions in the area of data management for high-performance and scientific computing. Attendees were challenged to identify promising approaches that would support the breadth of the DOE mission, including the explosion of artificial intelligence (AI) uses and the growing needs of experimental and observational science. Technological and science drivers were identified and considered as they relate to key aspects of data management such as interfaces, architectural design, and FAIR principles (Findable, Accessible, Interoperable, and Reusable). The thoughts of the workshop participants were distilled into a set of four priority research directions with the potential for high impact on DOE science. These research directions are summarized in the following pages.

96 KNOWLEDGE MANAGEMENT AND PRESERVATION↗

Orchestrating Fault Prediction with Live Migration and Checkpointing

Checkpoint/Restart (C/R) is widely used to provide fault tolerance on High-Performance Computing (HPC) systems. However, Parallel File System (PFS) overhead and failure uncertainty cause significant application overhead. This paper develops an adaptive multi-level C/R model that incorporates a failure prediction and analysis model, which orchestrates failure prediction, checkpointing, checkpoint frequency, and proactive live migration along with the additional benefit of Burst Buffers (BB). It effectively reduces the overheads due to failures, checkpointing, and recovery. Simulation results for the Summit supercomputer yield a reduction of ~20%-86% in application overhead due to BBs, orchestrated failure prediction, and migration. We also observe a ~29% decrease in checkpoint writes to BBs, which can increase the longevity of the BB storage devices.

Behera, Subhendu↗

Cooperative storage of shared files in a parallel computing system with dynamic block size

Improved techniques are provided for parallel writing of data to a shared object in a parallel computing system. A method is provided for storing data generated by a plurality of parallel processes to a shared object in a parallel computing system. The method is performed by at least one of the processes and comprises: dynamically determining a block size for storing the data; exchanging a determined amount of the data with at least one additional process to achieve a block of the data having the dynamically determined block size; and writing the block of the data having the dynamically determined block size to a file system. The determined block size comprises, e.g., a total amount of the data to be stored divided by the number of parallel processes. The file system comprises, for example, a log structured virtual parallel file system, such as a Parallel Log-Structured File System (PLFS).

97 MATHEMATICS AND COMPUTING↗

Scalable I/O Systems via Node-Local Storage: Approaching 1 TB/sec File I/O

The growth in the computational capability of modern supercomputing systems has been accompanied by corresponding increases in CPU count, total RAM, and total storage capacity. Indeed, systems such as Blue-Gene/L [3], BlueGene/P, Ranger, and the Cray XT series have grown to more than 100k processors, with 100 TeraBytes of RAM and are attached to multi-PetaByte storage systems. However, as part of this design evolution, large supercomputers have lost node-local storage elements, such as disks. While this decision was motivated by important considerations like overall system reliability, it also resulted in these systems losing a key level in their memory hierarchy, with nothing to fill the gap between local RAM and the parallel file system. While today's large supercomputers are typically attached to fast parallel file systems, which provide tens of GBs/s of I/O bandwidth, the computational capacity has grown much faster than the storage bandwidth capacity. As such, these machines are now provided with much less than 1GB/s of I/O bandwidth per TeraFlop of compute power, which is below the generally accepted limit required for a well-balanced system [8] [16]. The result is that today's limited I/O bandwidth is choking the capabilities of modern supercomputers, specifically in terms of limiting their working sets and making fault tolerance techniques, such as checkpointing, prohibitively expensive. This paper presents an alternative system design oriented on using node-local storage to improve aggregate system I/O bandwidth. We focus on the checkpointing use-case and present an experimental evaluation of SCR, a new checkpointing library that makes use of node-local storage to significantly improve the performance of checkpointing on large-scale supercomputers. Experiments show that SCR achieves unprecedented write speeds, reaching 700GB/s on 8,752 processors. Our results scale such that we expect a similarly structured system consisting of 12,500 processors to achieve aggregate I/O bandwidth of 1 TB/s.

99 GENERAL AND MISCELLANEOUS↗

Asynchronous Checkpoint Migration with MRNet in the Scalable Checkpoint / Restart Library

Applications running on today's supercomputers tolerate failures by periodically saving their state in checkpoint files on stable storage, such as a parallel file system. Although this approach is simple, the overhead of writing the checkpoints can be prohibitive, especially for large-scale jobs. In this paper, we present initial results of an enhancement to our Scalable Checkpoint/Restart Library (SCR). We employ MRNet, a tree-based overlay network library, to transfer checkpoints from the compute nodes to the parallel file system asynchronously. This enhancement increases application efficiency by removing the need for an application to block while checkpoints are transferred to the parallel file system. We show that the integration of SCR with MRNet can reduce the time spent in I/O operations by as much as 15x. However, our experiments exposed new scalability issues with our initial implementation. We discuss the sources of the scalability problems and our plans to address them.

97 MATHEMATICS AND COMPUTING↗

Integration experiences and performance studies of A COTS parallel archive systems

Current and future Archive Storage Systems have been asked to (a) scale to very high bandwidths, (b) scale in metadata performance, (c) support policy-based hierarchical storage management capability, (d) scale in supporting changing needs of very large data sets, (e) support standard interface, and (f) utilize commercial-off-the-shelf(COTS) hardware. Parallel file systems have been asked to do the same thing but at one or more orders of magnitude faster in performance. Archive systems continue to move closer to file systems in their design due to the need for speed and bandwidth, especially metadata searching speeds such as more caching and less robust semantics. Currently the number of extreme highly scalable parallel archive solutions is very small especially those that will move a single large striped parallel disk file onto many tapes in parallel. We believe that a hybrid storage approach of using COTS components and innovative software technology can bring new capabilities into a production environment for the HPC community much faster than the approach of creating and maintaining a complete end-to-end unique parallel archive software solution. In this paper, we relay our experience of integrating a global parallel file system and a standard backup/archive product with a very small amount of additional code to provide a scalable, parallel archive. Our solution has a high degree of overlap with current parallel archive products including (a) doing parallel movement to/from tape for a single large parallel file, (b) hierarchical storage management, (c) ILM features, (d) high volume (non-single parallel file) archives for backup/archive/content management, and (e) leveraging all free file movement tools in Linux such as copy, move, ls, tar, etc. We have successfully applied our working COTS Parallel Archive System to the current world's first petaflop/s computing system, LANL's Roadrunner, and demonstrated its capability to address requirements of future archival storage systems.

99 GENERAL AND MISCELLANEOUS↗

Integration experiments and performance studies of a COTS parallel archive system

Current and future Archive Storage Systems have been asked to (a) scale to very high bandwidths, (b) scale in metadata performance, (c) support policy-based hierarchical storage management capability, (d) scale in supporting changing needs of very large data sets, (e) support standard interface, and (f) utilize commercial-off-the-shelf (COTS) hardware. Parallel file systems have been asked to do the same thing but at one or more orders of magnitude faster in performance. Archive systems continue to move closer to file systems in their design due to the need for speed and bandwidth, especially metadata searching speeds such as more caching and less robust semantics. Currently the number of extreme highly scalable parallel archive solutions is very small especially those that will move a single large striped parallel disk file onto many tapes in parallel. We believe that a hybrid storage approach of using COTS components and innovative software technology can bring new capabilities into a production environment for the HPC community much faster than the approach of creating and maintaining a complete end-to-end unique parallel archive software solution. In this paper, we relay our experience of integrating a global parallel file system and a standard backup/archive product with a very small amount of additional code to provide a scalable, parallel archive. Our solution has a high degree of overlap with current parallel archive products including (a) doing parallel movement to/from tape for a single large parallel file, (b) hierarchical storage management, (c) ILM features, (d) high volume (non-single parallel file) archives for backup/archive/content management, and (e) leveraging all free file movement tools in Linux such as copy, move, Is, tar, etc. We have successfully applied our working COTS Parallel Archive System to the current world's first petafiop/s computing system, LANL's Roadrunner machine, and demonstrated its capability to address requirements of future archival storage systems.

99 GENERAL AND MISCELLANEOUS↗