Supporting computational data model representation with high-performance I/O in parallel netCDF

Abstract

Abstract-Parallel computational scientific applications have been described by their computation and communication patterns. From a storage and I/O perspective, these applications can also be grouped into separate data models based on the way data is organized and accessed during simulation, analysis, and visualization. Parallel netCDF is a popular library used in many scientific applications to store scientific datasets and provides high-performance parallel I/O. Although the metadata-rich netCDF file format can effectively store and describe regular multi-dimensional array datasets, it does not address the full range of current and future computational science data models. In this paper, we present a new storage scheme in Parallel netCDF to represent a broad variety of data models used in modern computational scientific applications. This scheme also allows concurrent metadata construction for different data objects from multiple groups of application processes, an important feature in obtaining a high degree of I/O parallelism for data models exhibiting irregular data distribution. Furthermore, we employ non-blocking I/O functions to aggregate irregularly distributed data requests into large, contiguous data requests, to achieve high-performance I/O. Using an example of adaptive mesh refinement data model, we demonstrate the proposed scheme can produce scalable performance results for both data and metadata creation and access.

An Asynchronous Parallel I/O Framework for Mass Conservation Ocean Model

“…There are several I/O methods based on NetCDF that are widely used in hydrodynamic models, as shown in Figure 1 NetCDF is a general basic library that defines a set of I/O functions and a machineindependent file format to support the creation, access, and sharing of array-oriented scientific data [5,6]. It is widely used by many physical simulation models, such as POP (Parallel Ocean Program) [7], CLM (Community Land Model) [8], CICE (Community Ice CodE) [9], and WRF (Weather Research and Forecasting) [10].…”

“…The collective method in PnetCDF can improve parallel I/O performance by significantly reducing many small and non-contiguous I/O requests [11]. PnetCDF supports irregular data distribution [5], MPI derived data types, and non-continuous data accessing from different processes on a single NetCDF file [11]. PnetCDF can collect multiple I/O requests over a record variable and optimize the I/O over a large pool of data transfers by producing more contiguous and larger transfers [15].…”

“…The equation for solving the vertical velocity ω in the p coordinate system is shown in Equation (3), and ∇ p represents the generalized derivative vector operator in the pressure vertical coordinate. The equations for solving the active tracers (temperature T and salinity S) are shown in Equations ( 4) and (5), where G hdi f f represents the parameterization of small-scale physics for horizontal tracer diffusion, G vdi f f represents the parameterization of small-scale physics for vertical tracer diffusion.…”

Investigating Read Performance of Python and NetCDF When Using HPC Parallel Filesystems

“…A solution for high speed, efficient I/O is to use a parallel data model, such as parallel-netcdf [15], or MPI-IO. The focus here is on what effect the nonparallel libraries have on the performance on a parallel HPC file system, so as to advise scientists on better read sizes and patterns to improve design of parallel data analysis code, therefore the parallel libraries were not used.…”

“…This will become particularly prevalent with the sixth community model inter-comparison project (CMIP6 [4]) expected to produce 10-20 PB of data [14]. This volume of data means that high performance I/O is a critical requirement for efficient analysis code, and therefore productive scientific workflows [15].…”