Search

The applications being developed within the U.S. Exascale Computing Project (ECP) to run on imminent Exascale computers will generate scientific results with unprecedented fidelity and record turn-around time. Many of these codes are based on particle-mesh methods and use advanced algorithms, especially dynamic load-balancing and mesh-refinement, to achieve high performance on Exascale machines. Yet, as such algorithms improve parallel application efficiency, they raise new challenges for I/O logic due to their irregular and dynamic data distributions. Thus, while the enormous data rates of Exascale simulations already challenge existing file system write strategies, the need for efficient read and processing of generated data introduces additional constraints on the data layout strategies that can be used when writing data to secondary storage. We review these I/O challenges and introduce two online data layout reorganization approaches for achieving good tradeoffs between read and write performance. We demonstrate the benefits of using these two approaches for the ECP particle-in-cell simulation WarpX, which serves as a motif for a large class of important Exascale applications. We show that by understanding application I/O patterns and carefully designing data layouts we can increase read performance by more than 80%., 12 pages, 15 figures, accepted by IEEE Transactions on Parallel and Distributed Systems

Author(s): Godoy, WF; Podhorszki, N; Wang, R; Atkins, C; Eisenhauer, G; Gu, J; Davis, P; Choi, J; Germaschewski, K; Huck, K; Huebl, A; Kim, M; Kress, J; Kurc, T; Liu, Q; Logan, J; Mehta, K; Ostrouchov, G; Parashar, M; Poeschel, F; Pugmire, D; Suchyta, E; Takahashi, K; Thompson, N; Tsutsumi, S; Wan, L; Wolf, M; Wu, K; Klasky, S | Abstract: We present ADIOS 2, the latest version of the Adaptable Input Output (I/O) System. ADIOS 2 addresses scientific data management needs ranging from scalable I/O in supercomputers, to data analysis in personal computer and cloud systems. Version 2 introduces a unified application programming interface (API) that enables seamless data movement through files, wide-area-networks, and direct memory access, as well as high-level APIs for data analysis. The internal architecture provides a set of reusable and extendable components for managing data presentation and transport mechanisms for new applications. ADIOS 2 bindings are available in C++11, C, Fortran, Python, and Matlab and are currently used across different scientific communities. ADIOS 2 provides a communal framework to tackle data management challenges as we approach the exascale era of supercomputing.

Upcoming exascale applications could introduce significant data management challenges due to their large sizes, dynamic work distribution, and involvement of accelerators such as graphical processing units, GPUs. In this work, we explore the performance of reading and writing operations involving one such scientific application on two different supercomputers. Our tests showed that the Adaptable Input and Output System, ADIOS, was able to achieve speeds over 1TB/s, a significant fraction of the peak I/O performance on Summit. We also demonstrated the querying functionality in ADIOS could effectively support common selective data analysis operations, such as conditional histograms. In tests, this query mechanism was able to reduce the execution time by a factor of five. More importantly, ADIOS data management framework allows us to achieve these performance improvements with only a minimal amount of coding effort.

Author(s): O'Brien, TA; Collins, WD; Kashinath, K; Rubel, O; Byna, S; Gu, J; Krishnan, H; Ullrich, PA | Abstract: Numerous studies have shown that atmospheric models with high horizontal resolution better represent the physics and statistics of precipitation in climate models. While it is abundantly clear from these studies that high-resolution increases the rate of extreme precipitation, it is not clear whether these added extreme events are “realistic”; whether they occur in simulations in response to the same forcings that drive similar events in reality. In order to understand whether increasing horizontal resolution results in improved model fidelity, a hindcast-based, multiresolution experimental design has been conceived and implemented: the InitiaLIzed-ensemble, Analyze, and Develop (ILIAD) framework. The ILIAD framework allows direct comparison between observed and simulated weather events across multiple resolutions and assessment of the degree to which increased resolution improves the fidelity of extremes. Analysis of 5 years of daily 5 day hindcasts with the Community Earth System Model at horizontal resolutions of 220, 110, and 28 km shows that: (1) these hindcasts reproduce the resolution-dependent increase of extreme precipitation that has been identified in longer-duration simulations, (2) the correspondence between simulated and observed extreme precipitation improves as resolution increases; and (3) this increase in extremes and precipitation fidelity comes entirely from resolved-scale precipitation. Evidence is presented that this resolution-dependent increase in precipitation intensity can be explained by the theory of Rauscher et al. (), which states that precipitation intensifies at high resolution due to an interaction between the emergent scaling (spectral) properties of the wind field and the constraint of fluid continuity.

Scientific discoveries are increasingly relying on analysis of massive amounts of data. The ability to directly access the most relevant data records through query, without shifting through all of them becomes essential. However, scientific datasets are commonly stored on parallel file systems and I/O systems that are optimized for reading/writing large chunks of data, and many scientific datasets have spatial-temporal data similarity, such that the records with similar values often locate in a close proximity of each other. Therefore, our previous work started to investigate the benefit of using block range index technique for scientific datasets, which only records the value range of all the records in a data block. In this paper, we extend our work in several aspects. First, we implement and integrate our block index technique with the ADIOS I/O system. Second, we show our proposed method can be significantly better than the existing minmax and bitmaps indexing methods supported in ADIOS, and can also have comparable performance in the worst case. Third, we propose several techniques that can take advantage of the block index information to greatly reduce data retrieval time from query results. Fourth, we evaluate our approach using several real scientific datasets, and analyze the spatial-temporal data similarity characteristics in them. Through our study, we believe block index can be an effective indexing technique for scientific datasets with little implementation and operating overhead. It's size is small enough for building the indexes on-the-fly, and yet its query information is sufficient for efficient data access.

© 2016 IEEE. Data driven science is becoming increasingly more common, complex, and is placing tremendous stresses on visualization and analysis frameworks. Data sources producing 10GB per second (and more) are becoming increasingly commonplace in both simulation, sensor and experimental sciences. These data sources, which are often distributed around the world, must be analyzed by teams of scientists that are also distributed. Enabling scientists to view, query and interact with such large volumes of data in near-real-time requires a rich fusion of visualization and analysis techniques, middleware and workflow systems. This paper discusses initial research into visualization and analysis of distributed data workflows that enables scientists to make near-real-time decisions of large volumes of time varying data.

Before scientific analyses run on shared infrastructure, such as the Open Science Grid or XSEDE, scientists must often transfer or stage key data sets those resources. Often these datasets consist of many files that may be transferred by multiple clients in parallel. We study two techniques that improve the use of available resources for these large, long-running, multi-file transfers. First, we adapt transfer parameters for multi-file transfers based on recent transfer performance. Second, we use VO and site policies to influence the allocation of system resources for transfers, such as available transfer streams. We describe our system design and summarize its implementation and performance.

With scientific data collected at unprecedented volumes and rates, the success of large scientific collaborations requires that they provide distributed data access with improved data access latencies and increased reliability to a large user community. The goal of the ADAPT (Adaptive Data Access and Policy-driven Transfers) project is to develop and deploy a general-purpose data access framework for scientific collaborations that provides fine-grained and adaptive data transfer management and the use of site and VO policies for resource sharing. This paper presents our initial design and implementation of an adaptive data transfer framework. We also present preliminary performance measurements showing that adaptation and policy improve network performance.

Modern scientific data-intensive applications brought about the need for novel data transfer technologies and automated tools capable of effectively utilizing available raw network bandwidth and intelligently assisting scientists in replicating large volumes of data to desired locations in a timely manner. In this paper we describe the design of StorNet, an integrated end-to-end resource provisioning and management system for high performance data transfers that can operate with heterogeneous network protocols and storage systems in a federated computing environment. StorNet allocates and co-schedules storage and network resources involved in data transfers. It is based on existing Storage Resource Manager, TeraPaths, and OSCARS capabilities. StorNet provides data intensive applications with the capability of predictable, yet efficient delivery of data at rates of multiple gigabits/second, bridging end-to-end advanced storage and network technologies in a transparent way.

The amount of scientific data generated by simulations orcollected from large scale experiments have reached levels that cannot bestored in the researcher's workstation or even in his/her local computercenter. Such data are vital to large scientific collaborations dispersedover wide-area networks. In the past, the concept of a Gridinfrastructure [1]mainly emphasized the computational aspect ofsupporting large distributed computational tasks, and optimizing the useof the network by using bandwidth reservation techniques. In this paperwe discuss the concept of Storage Resource Managers (SRMs) as componentsthat complement this with the support for the storage management of largedistributed datasets. The access to data is becoming the main bottleneckin such "data intensive" applications because the data cannot bereplicated in all sites. SRMs can be used to dynamically optimize the useof storage resource to help unclog this bottleneck.

Storage Resource Managers (SRMs) are middleware components whose function is to provide dynamic space allocation and file management of shared storage components on the Grid. They complement Compute Resource Managers and Network Resource Managers in providing storage reservation and dynamic information on storage availability for the planning and execution of a Grid job. SRMs manage two types of resources: space and files. When managing space, SRMs negotiate space allocation with the requesting client, and/or assign default space quotas. When managing files, SRMs allocate space for files, invoke file transfer services to move files into the space, pin files for a certain lifetime, release files upon the clients request, and use file replacement policies to optimize the use of the shared space. SRMs can be designed to provide effective sharing of files, by monitoring the activity of shared files, and make dynamic decisions on which files to replace when space is needed. In addition, SRMs perform automatic garbage collection of unused files by removing selected files whose lifetime has expired when space is needed. In this chapter we discuss the design considerations for SRMs, their functionality, and their interfaces. We demonstrate the use of SRMs with several examples of real implementations that are in use today in a routine fashion or in a prototype form.

High Energy Nuclear Physics (HENP) experiments such as STAR at BNL and ATLAS at CERN produce large amounts of data that are stored as files on mass storage systems in computer centers. In these files, the basic unit of data is an event. Analysis is typically performed on a selected set of events. The files containing these events have to be located, copied from mass storage systems to disks before analysis, and removed when no longer needed. These file management tasks are tedious and time consuming. Typically, all events contained in the files are read into memory before a selection is made. Since the time to read the events dominate the overall execution time, reading the unwanted event needlessly increases the analysis time. The Grid Collector is a set of software modules that works together to address these two issues. It automates the file management tasks and provides "direct" access to the selected events for analyses. It is currently integrated with the STAR analysis framework. The users can select events based on tags, such as, "production date between March 10 and 20, and the number of charged tracks > 100." The Grid Collector locates the files containing relevant events, transfers the files across the Grid if necessary, and delivers the events to the analysis code through the familiar iterators. There has been some research efforts to address the file management issues, the Grid Collector is unique in that it addresses the event access issue together with the file management issues. This makes it more useful to a large variety of users.

StorNet is a joint project of Brookhaven National Laboratory (BNL) and Lawrence Berkeley National Laboratory (LBNL) to research, design, and develop an integrated end-to-end resource provisioning and management framework for high-performance data transfers. The StorNet framework leverages heterogeneous network protocols and storage types in a federated computing environment to provide the capability of predictable, efficient delivery of high-bandwidth data transfers for data intensive applications. The framework incorporates functional modules to perform such data transfers through storage and network bandwidth co-scheduling, storage and network resource provisioning, and performance monitoring, and is based on LBNL's BeStMan/SRM, BNL's TeraPaths, and ESNet's OSCARS systems.

As scientific instruments and computer simulations produce more and more data, the task of locating the essential information to gain insight becomes increasingly difficult. FastBit is an efficient software tool to address this challenge. In this article, we present a summary of the key underlying technologies, namely bitmap compression, encoding, and binning. Together these techniques enable FastBit to answer structured (SQL) queries orders of magnitude faster than popular database systems. To illustrate how FastBit is used in applications, we present three examples involving a high-energy physics experiment, a combustion simulation, and an accelerator simulation. In each case, FastBit significantly reduces the response time and enables interactive exploration on terabytes of data.

Facing the reality of storage economics, NP experiments such as RHIC/STAR have been engaged in a shift of the analysis model, and now heavily rely on using cheap disks attached to processing nodes, as such a model is extremely beneficial over expensive centralized storage. Additionally, exploiting storage aggregates with enhanced distributed computing capabilities such as dynamic space allocation (lifetime of spaces), file management on shared storages (lifetime of files, pinning file), storage policies or a uniform access to heterogeneous storage solutions is not an easy task. The Xrootd/Scalla system allows for storage aggregation. We will present an overview of the largest deployment of Scalla (Structured Cluster Architecture for Low Latency Access) in the world spanning over 1000 CPUs co-sharing the 350 TB Storage Elements and the experience on how to make such a model work in the RHIC/STAR standard analysis framework. We will explain the key features and approach on how to make access to mass storage (HPSS) possible in such a large deployment context. Furthermore, we will give an overview of a fully 'gridified' solution using the plug-and-play features of Scalla architecture, replacing standard storage access with grid middleware SRM (Storage Resource Manager) components designed for space management and will compare the solution with the standard Scalla approach in use in STAR for the past 2 years. Integration details, future plans and status of development will be explained in the area of best transfer strategy between multiple-choice data pools and best placement with respect of load balancing and interoperability with other SRM aware tools or implementations.

Storage Services are crucial components of the Worldwide LHC Computing Grid Infrastructure spanning more than 200 sites and serving computing and storage resources to the High Energy Physics LHC communities. Up to tens of Petabytes of data are collected every year by the four LHC experiments at CERN. To process these large data volumes it is important to establish a protocol and a very efficient interface to the various storage solutions adopted by the WLCG sites. In this work we report on the experience acquired during the definition of the Storage Resource Manager v2.2 protocol. In particular, we focus on the study performed to enhance the interface and make it suitable for use by the WLCG communities. At the moment 5 different storage solutions implement the SRM v2.2 interface: BeStMan (LBNL), CASTOR (CERN and RAL), dCache (DESY and FNAL), DPM (CERN), and StoRM (INFN and ICTP). After a detailed inside review of the protocol, various test suites have been written identifying the most effective set of tests: the S2 test suite from CERN and the SRM- Tester test suite from LBNL. Such test suites have helped verifying the consistency and coherence of the proposed protocol and validating existing implementations. We conclude our work describing the results achieved.