Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811892
N. Ali, A. Devulapalli, D. Dalessandro, P. Wyckoff, P. Sadayappan
As the types of problems we solve in high-performance computing and other areas become more complex, the amount of data generated and used is growing at a rapid rate. Today many terabytes of data are common; tomorrow petabytes of data will be the norm. Much work has been put into increasing capacity and I/O performance for large-scale storage systems. However, one often ignored area is metadata management. Metadata can have a significant impact on the performance of a system. Past approaches have moved metadata activities to a separate server in order to avoid potential interference with data operations. However, with the advent of object-based storage technology, there is a compelling argument to re-couple metadata and data. In this paper we present two metadata management schemes, both of which remove the need for a separate metadata server and replace it with object-based storage.
{"title":"Revisiting the metadata architecture of parallel file systems","authors":"N. Ali, A. Devulapalli, D. Dalessandro, P. Wyckoff, P. Sadayappan","doi":"10.1109/PDSW.2008.4811892","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811892","url":null,"abstract":"As the types of problems we solve in high-performance computing and other areas become more complex, the amount of data generated and used is growing at a rapid rate. Today many terabytes of data are common; tomorrow petabytes of data will be the norm. Much work has been put into increasing capacity and I/O performance for large-scale storage systems. However, one often ignored area is metadata management. Metadata can have a significant impact on the performance of a system. Past approaches have moved metadata activities to a separate server in order to avoid potential interference with data operations. However, with the advent of object-based storage technology, there is a compelling argument to re-couple metadata and data. In this paper we present two metadata management schemes, both of which remove the need for a separate metadata server and replace it with object-based storage.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134546361","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811890
M. Storer, K. Greenan, I. Adams, E. L. Miller, D. Long, K. Voruganti
Archival storage systems designed to preserve scientific data, business data, and consumer data must maintain and safeguard tens to hundreds of petabytes of data on tens of thousands of media for decades. Such systems are currently designed in the same way as higher-performance, shorter-term storage systems, which have a useful lifetime but must be replaced in their entirety via a ldquofork-liftrdquo upgrade. Thus, while existing solutions can provide good energy efficiency and relatively low cost, they do not adapt well to continuous improvements in technology, becoming less efficient relative to current technology as they age. In an archival storage environment, this paradigm implies an endless series of wholesale migrations and upgrades to remain efficient and up to date. Our approach, Logan, manages node addition, removal, and failure on a distributed network of intelligent storage appliances, allowing the system to gradually evolve as device technology advances. By automatically handling most of the common administration chores-integrating new devices into the system, managing groups of devices that work together to provide redundancy, and recovering from failed devices-Logan reduces management overhead and thus cost. Logan can also improve cost and space efficiency by identifying and decommissioning outdated devices, thus reducing space and power requirements for the archival storage system.
{"title":"Logan: Automatic management for evolvable, large-scale, archival storage","authors":"M. Storer, K. Greenan, I. Adams, E. L. Miller, D. Long, K. Voruganti","doi":"10.1109/PDSW.2008.4811890","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811890","url":null,"abstract":"Archival storage systems designed to preserve scientific data, business data, and consumer data must maintain and safeguard tens to hundreds of petabytes of data on tens of thousands of media for decades. Such systems are currently designed in the same way as higher-performance, shorter-term storage systems, which have a useful lifetime but must be replaced in their entirety via a ldquofork-liftrdquo upgrade. Thus, while existing solutions can provide good energy efficiency and relatively low cost, they do not adapt well to continuous improvements in technology, becoming less efficient relative to current technology as they age. In an archival storage environment, this paradigm implies an endless series of wholesale migrations and upgrades to remain efficient and up to date. Our approach, Logan, manages node addition, removal, and failure on a distributed network of intelligent storage appliances, allowing the system to gradually evolve as device technology advances. By automatically handling most of the common administration chores-integrating new devices into the system, managing groups of devices that work together to provide redundancy, and recovering from failed devices-Logan reduces management overhead and thus cost. Logan can also improve cost and space efficiency by identifying and decommissioning outdated devices, thus reducing space and power requirements for the archival storage system.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127859479","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811886
Milo Polte, J. Simsa, Garth A. Gibson
In terms of performance, solid state devices promise to be superior technology to mechanical disks. This study investigates performance of several up-to-date high-end consumer and enterprise Flash solid state devices (SSDs) and relates their performance to that of mechanical disks. For the purpose of this evaluation, the IOZone benchmark is run in single-threaded mode with varying request size and access pattern on an ext3 filesystem mounted on these devices. The price of the measured devices is then used to allow for comparison of price per performance. Measurements presented in this study offer an evaluation of cost-effectiveness of a Flash based SSD storage solution over a range of workloads. In particular, for sequential access pattern the SSDs are up to 10 times faster for reads and up to 5 times faster than the disks. For random reads, the SSDs provide up to 200times performance advantage. For random writes the SSDs provide up to 135times performance advantage. After weighting these numbers against the prices of the tested devices, we can conclude that SSDs are approaching price per performance of magnetic disks for sequential access patterns workloads and are superior technology to magnetic disks for random access patterns.
{"title":"Comparing performance of solid state devices and mechanical disks","authors":"Milo Polte, J. Simsa, Garth A. Gibson","doi":"10.1109/PDSW.2008.4811886","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811886","url":null,"abstract":"In terms of performance, solid state devices promise to be superior technology to mechanical disks. This study investigates performance of several up-to-date high-end consumer and enterprise Flash solid state devices (SSDs) and relates their performance to that of mechanical disks. For the purpose of this evaluation, the IOZone benchmark is run in single-threaded mode with varying request size and access pattern on an ext3 filesystem mounted on these devices. The price of the measured devices is then used to allow for comparison of price per performance. Measurements presented in this study offer an evaluation of cost-effectiveness of a Flash based SSD storage solution over a range of workloads. In particular, for sequential access pattern the SSDs are up to 10 times faster for reads and up to 5 times faster than the disks. For random reads, the SSDs provide up to 200times performance advantage. For random writes the SSDs provide up to 135times performance advantage. After weighting these numbers against the prices of the tested devices, we can conclude that SSDs are approaching price per performance of magnetic disks for sequential access patterns workloads and are superior technology to magnetic disks for random access patterns.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130827696","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811881
J. Lofstead, F. Zheng, S. Klasky, K. Schwan
Scientific Data Management has become essential to the productivity of scientists using ever larger machines and running applications that produce ever more data. There are several specific issues when running on petascale (and beyond) machines. One is the need for massively parallel data output, which in part, depends on the data formats and semantics being used. Here, the inhibition of parallelism by file system notions of strict and immediate consistency can be addressed with ldrdelayed data consistencypsila methods. Such methods can also be used to remove the runtime coordination steps required for immediate consistency from machine resources like Bluegene's separate networks for barrier calls and its dedicated IO nodes, thereby freeing them to instead, perform alternate tasks that enhance data output performance and/or richness. Second, once data is generated, it is important to be able to efficiently access it, which implies the need for rapid data characterization and indexing. This can be achieved by adding small amounts of metadata to the output process, thereby permitting scientists to quickly make informed decisions about which files to process from large-scale science runs. Third, failure probabilities increase with an increasing number of nodes, which suggests the need for organizing output data to be resilient to failures in which the output from a single or from a small number of nodes is lost or corrupted. This paper demonstrates the utility of using delayed consistency methods for the process of data output from the compute nodes of petascale machines. It also demonstrates the advantages derived from resilient data organization coupled with lightweight methods for data indexing. An implementation of these techniques is realized in ADIOS, the Adaptable IO System, and its BP intermediate file format. The implementation is designed to be compatible with existing, well-known file formats like HDF-5 and NetCDF, thereby permitting end users to exploit the rich tool chains for these formats. Initial performance evaluations of the approach exhibit substantial performance advantages over using native parallel HDF-5 in the Chimera supernova code.
科学数据管理对于使用越来越大的机器和运行产生越来越多数据的应用程序的科学家的生产力已经变得至关重要。在千兆级(甚至更高)机器上运行时有几个特定的问题。一个是需要大规模并行数据输出，这在一定程度上取决于所使用的数据格式和语义。在这里，文件系统严格和即时一致性概念对并行性的抑制可以用ldrdelayed data consistencsila方法解决。这些方法还可以用于从机器资源中移除即时一致性所需的运行时协调步骤，例如Bluegene的屏障调用和专用IO节点的单独网络，从而释放它们来执行增强数据输出性能和/或丰富性的替代任务。其次，一旦生成数据，重要的是能够有效地访问它，这意味着需要快速的数据表征和索引。这可以通过向输出过程中添加少量元数据来实现，从而允许科学家快速做出明智的决定，决定从大规模科学运行中处理哪些文件。第三，故障概率随着节点数量的增加而增加，这表明需要组织输出数据以适应单个或少数节点的输出丢失或损坏的故障。本文演示了使用延迟一致性方法处理千万亿级计算机计算节点的数据输出过程的实用性。它还演示了弹性数据组织与轻量级数据索引方法相结合所带来的优势。这些技术在ADIOS (adaptive IO System)及其BP中间文件格式中实现。该实现旨在与现有的知名文件格式(如HDF-5和NetCDF)兼容，从而允许最终用户利用这些格式的丰富工具链。该方法的初步性能评估显示，与在Chimera超新星代码中使用本地并行HDF-5相比，该方法具有显著的性能优势。
{"title":"Input/output APIs and data organization for high performance scientific computing","authors":"J. Lofstead, F. Zheng, S. Klasky, K. Schwan","doi":"10.1109/PDSW.2008.4811881","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811881","url":null,"abstract":"Scientific Data Management has become essential to the productivity of scientists using ever larger machines and running applications that produce ever more data. There are several specific issues when running on petascale (and beyond) machines. One is the need for massively parallel data output, which in part, depends on the data formats and semantics being used. Here, the inhibition of parallelism by file system notions of strict and immediate consistency can be addressed with ldrdelayed data consistencypsila methods. Such methods can also be used to remove the runtime coordination steps required for immediate consistency from machine resources like Bluegene's separate networks for barrier calls and its dedicated IO nodes, thereby freeing them to instead, perform alternate tasks that enhance data output performance and/or richness. Second, once data is generated, it is important to be able to efficiently access it, which implies the need for rapid data characterization and indexing. This can be achieved by adding small amounts of metadata to the output process, thereby permitting scientists to quickly make informed decisions about which files to process from large-scale science runs. Third, failure probabilities increase with an increasing number of nodes, which suggests the need for organizing output data to be resilient to failures in which the output from a single or from a small number of nodes is lost or corrupted. This paper demonstrates the utility of using delayed consistency methods for the process of data output from the compute nodes of petascale machines. It also demonstrates the advantages derived from resilient data organization coupled with lightweight methods for data indexing. An implementation of these techniques is realized in ADIOS, the Adaptable IO System, and its BP intermediate file format. The implementation is designed to be compatible with existing, well-known file formats like HDF-5 and NetCDF, thereby permitting end users to exploit the rich tool chains for these formats. Initial performance evaluations of the approach exhibit substantial performance advantages over using native parallel HDF-5 in the Chimera supernova code.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133143641","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811884
A. Leung, E. L. Miller
As file system capacities reach the petascale, it is becoming increasingly difficult for users to organize, find, and manage their data. File system search has the potential to greatly improve how users manage and access files. Unfortunately, existing file system search is designed for smaller scale systems, making it difficult for existing solutions to scale to petascale files systems. In this paper, we motivate the importance of file system search in petascale file systems and present a new full text file system search design for petascale file systems. Unlike existing solutions, our design exploits file system properties. Using a novel index partitioning mechanism that utilizes file system namespace locality, we are able to improve search scalability and performance and we discuss how such a design can potentially improve search security and ranking.We describe how our design can be implemented within the Ceph petascale file system.
{"title":"Scalable full-text search for petascale file systems","authors":"A. Leung, E. L. Miller","doi":"10.1109/PDSW.2008.4811884","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811884","url":null,"abstract":"As file system capacities reach the petascale, it is becoming increasingly difficult for users to organize, find, and manage their data. File system search has the potential to greatly improve how users manage and access files. Unfortunately, existing file system search is designed for smaller scale systems, making it difficult for existing solutions to scale to petascale files systems. In this paper, we motivate the importance of file system search in petascale file systems and present a new full text file system search design for petascale file systems. Unlike existing solutions, our design exploits file system properties. Using a novel index partitioning mechanism that utilizes file system namespace locality, we are able to improve search scalability and performance and we discuss how such a design can potentially improve search security and ranking.We describe how our design can be implemented within the Ceph petascale file system.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124937299","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811889
Grant Mackey, S. Sehrish, John Bent, J. López, S. Habib, J. Wang
In this work we present an scientific application that has been given a Hadoop MapReduce implementation. We also discuss other scientific fields of supercomputing that could benefit from a MapReduce implementation. We recognize in this work that Hadoop has potential benefit for more applications than simply data mining, but that it is not a panacea for all data intensive applications. We provide an example of how the halo finding application, when applied to large astrophysics datasets, benefits from the model of the Hadoop architecture. The halo finding application uses a friends of friends algorithm to quickly cluster together large sets of particles to output files which a visualization software can interpret. The current implementation requires that large datasets be moved from storage to computation resources for every simulation of astronomy data. Our Hadoop implementation allows for an in-place halo finding application on the datasets, which removes the time consuming process of transferring data between resources.
{"title":"Introducing map-reduce to high end computing","authors":"Grant Mackey, S. Sehrish, John Bent, J. López, S. Habib, J. Wang","doi":"10.1109/PDSW.2008.4811889","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811889","url":null,"abstract":"In this work we present an scientific application that has been given a Hadoop MapReduce implementation. We also discuss other scientific fields of supercomputing that could benefit from a MapReduce implementation. We recognize in this work that Hadoop has potential benefit for more applications than simply data mining, but that it is not a panacea for all data intensive applications. We provide an example of how the halo finding application, when applied to large astrophysics datasets, benefits from the model of the Hadoop architecture. The halo finding application uses a friends of friends algorithm to quickly cluster together large sets of particles to output files which a visualization software can interpret. The current implementation requires that large datasets be moved from storage to computation resources for every simulation of astronomy data. Our Hadoop implementation allows for an in-place halo finding application on the datasets, which removes the time consuming process of transferring data between resources.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"332 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124679709","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811887
M. Curry, A. Skjellum, H. Ward, R. Brightwell
Reed-Solomon coding is a method of generating arbitrary amounts of checksum information from original data via matrix-vector multiplication in finite fields. Previous work has shown that CPUs are not well-matched to this type of computation, but recent graphical processing units (GPUs) have been shown through a case study to perform this encoding quickly for the 3 + 3 (three data + three parity) case. In order to be utilized in a true RAID-like system, it is important to understand how well this computation can scale in the number of data disks supported. This paper details the performance of a general Reed-Solomon encoding and decoding library that is suitable for use in RAID-like systems. Both generation and recovery are performance-tested and discussed.
{"title":"Arbitrary dimension Reed-Solomon coding and decoding for extended RAID on GPUs","authors":"M. Curry, A. Skjellum, H. Ward, R. Brightwell","doi":"10.1109/PDSW.2008.4811887","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811887","url":null,"abstract":"Reed-Solomon coding is a method of generating arbitrary amounts of checksum information from original data via matrix-vector multiplication in finite fields. Previous work has shown that CPUs are not well-matched to this type of computation, but recent graphical processing units (GPUs) have been shown through a case study to perform this encoding quickly for the 3 + 3 (three data + three parity) case. In order to be utilized in a true RAID-like system, it is important to understand how well this computation can scale in the number of data disks supported. This paper details the performance of a general Reed-Solomon encoding and decoding library that is suitable for use in RAID-like systems. Both generation and recovery are performance-tested and discussed.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129423769","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This report describes how a file system level log-based technique can improve the write performance of many-to-one write checkpoint workload typical for high performance computations. It is shown that a simple log-based organization can provide for substantial improvements in the write performance while retaining the convenience of a single flat file abstraction. The improvement of the write performance comes at the cost of degraded read performance however. Techniques to alleviate the read performance penalty, such as file reconstruction on the first read, are discussed.
{"title":"Fast log-based concurrent writing of checkpoints","authors":"Milo Polte, Jiri Simsa, Wittawat Tantisiriroj, Garth A. Gibson, Shobhit Dayal, Mikhail Chainani, Dilip Kumar Uppugandla","doi":"10.1109/PDSW.2008.4811882","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811882","url":null,"abstract":"This report describes how a file system level log-based technique can improve the write performance of many-to-one write checkpoint workload typical for high performance computations. It is shown that a simple log-based organization can provide for substantial improvements in the write performance while retaining the convenience of a single flat file abstraction. The improvement of the write performance comes at the cost of degraded read performance however. Techniques to alleviate the read performance penalty, such as file reconstruction on the first read, are discussed.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129366709","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811883
P. Nowoczynski, N. Stone, J. Yanovich, J. Sommerfield
The PSC has developed a prototype distributed file system infrastructure that vastly accelerates aggregated write bandwidth on large compute platforms. Write bandwidth, more than read bandwidth, is the dominant bottleneck in HPC I/O scenarios due to writing checkpoint data, visualization data and post-processing (multi-stage) data. We have prototyped a scalable solution that will be directly applicable to future petascale compute platforms having of order 10^6 cores. Our design emphasizes high-efficiency scalability, low-cost commodity components, lightweight software layers, end-to-end parallelism, client-side caching and software parity, and a unique model of load-balancing outgoing I/O onto high-speed intermediate storage followed by asynchronous reconstruction to a 3rd-party parallel file system.
{"title":"Zest Checkpoint storage system for large supercomputers","authors":"P. Nowoczynski, N. Stone, J. Yanovich, J. Sommerfield","doi":"10.1109/PDSW.2008.4811883","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811883","url":null,"abstract":"The PSC has developed a prototype distributed file system infrastructure that vastly accelerates aggregated write bandwidth on large compute platforms. Write bandwidth, more than read bandwidth, is the dominant bottleneck in HPC I/O scenarios due to writing checkpoint data, visualization data and post-processing (multi-stage) data. We have prototyped a scalable solution that will be directly applicable to future petascale compute platforms having of order 10^6 cores. Our design emphasizes high-efficiency scalability, low-cost commodity components, lightweight software layers, end-to-end parallelism, client-side caching and software parity, and a unique model of load-balancing outgoing I/O onto high-speed intermediate storage followed by asynchronous reconstruction to a 3rd-party parallel file system.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131564676","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2008-11-01DOI: 10.1109/PDSW.2008.4811891
H. M. Monti, A. Butt, S. Vazhkudai
High performance computing is facing a data deluge from state-of-the-art colliders and observatories. Large data-sets from these facilities, and other end-user sites, are often inputs to intensive analyses on modern supercomputers. Timely staging in of input data at the supercomputer's local storage can not only optimize space usage, but also protect against delays due to storage system failures. To this end, we propose a just-in-time staging framework that uses a combination of batch-queue predictions, user-specified intermediate nodes, and decentralized data delivery to coincide input data staging with job startup. Our preliminary prototype has been integrated with widely used tools such as the PBS job submission system, BitTorrent data delivery, and Network Weather Service network monitoring facility.
{"title":"Just-in-time staging of large input data for supercomputing jobs","authors":"H. M. Monti, A. Butt, S. Vazhkudai","doi":"10.1109/PDSW.2008.4811891","DOIUrl":"https://doi.org/10.1109/PDSW.2008.4811891","url":null,"abstract":"High performance computing is facing a data deluge from state-of-the-art colliders and observatories. Large data-sets from these facilities, and other end-user sites, are often inputs to intensive analyses on modern supercomputers. Timely staging in of input data at the supercomputer's local storage can not only optimize space usage, but also protect against delays due to storage system failures. To this end, we propose a just-in-time staging framework that uses a combination of batch-queue predictions, user-specified intermediate nodes, and decentralized data delivery to coincide input data staging with job startup. Our preliminary prototype has been integrated with widely used tools such as the PBS job submission system, BitTorrent data delivery, and Network Weather Service network monitoring facility.","PeriodicalId":227342,"journal":{"name":"2008 3rd Petascale Data Storage Workshop","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2008-11-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122380106","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：