Skip to main content

Eagle System Configuration

 Learn about the Eagle system configuration.

Architecture Description

The Eagle system is a high-performance computing (HPC) system with different types of servers (nodes) configured to run compute intensive and parallel computing jobs. All nodes run the Linux operating system: Red Hat Linux or the derivative CentOS distribution. The nodes and storage are connected by a high-speed 100 Gb/sec EDR InfiniBand network. A brief description of the configuration and features of the nodes, interconnect and file systems is provided below. 

Compute Node Hardware Details

Eagle has 2100 compute nodes available for HPC jobs. Additional nodes maybe available, but are not guaranteed. A variety of different node types are available:

Number of Nodes Memory Processors Accelerators Local Storage
1728 96 GB Dual Intel Xeon Gold Skylake 6154 (3.0 GHz, 18-core) processors N/A 1 TB SATA
288 192 GB Dual Intel Xeon Gold Skylake 6154 (3.0 GHz, 18-core) processors N/A 1 TB SATA
48 768 GB Dual Intel Xeon Gold Skylake 6154 (3.0 GHz, 18-core) processors N/A

10 nodes with 25.6 TB SSD

38 nodes with 1.6 TB SSD

50 768 GB Dual Intel Xeon Gold Skylake 6154 (3.0 GHz, 18-core) processors Dual NVIDIA Tesla V100 PCIe 16 GB Computational Accelerator

10 nodes with 25.6 TB SSD

40 nodes with 1.6 TB SSD

Login Nodes

There are three login nodes on the system.  The /home, /nopt, /scratch, /projects, /shared-projects, /datasets and /mss file systems are mounted on all login nodes. 

Users may connect to eagle.hpc.nrel.gov from the NREL network. This will connect to one of the three login nodes. Users also have the option of connecting directly to an individual login node using one of the following names: 

  • el1.hpc.nrel.gov
  • el2.hpc.nrel.gov
  • el3.hpc.nrel.gov

To connect to Eagle from outside the NREL network, use eagle.nrel.gov.

Data Analysis and Visualization Nodes

The Data Analysis & Visualization (DAV) nodes are each equipped with Dual Intel Xeon Gold Skylake 6154 (3.0 GHz, 18-core) processors and dual NVIDIA Tesla V100 PCIe 16 GB Computational Accelerators. These nodes support OpenCL and CUDA programming models. These nodes support hardware-accelerated remote visualization of data using the FastX remote desktop and visualization software.

Users may connect to ed.hpc.nrel.gov. This will connect to one of the three DAV nodes. Users also have the option of connecting directly to an individual DAV node using one of the following:

  • ed1.hpc.nrel.gov
  • ed2.hpc.nrel.gov
  • ed3.hpc.nrel.gov

To connect to Eagle DAV/FastX from outside the NREL network, use eagle-dav.nrel.gov.

Interconnect

All nodes and storage are connected using an enhanced 8-dimensional InfiniBand Enhanced Data Rate (EDR - 100 Gb/sec) hypercube topology that provides a bisection bandwidth of 26.4 TB/sec.

Home File System

The Home File System (HFS) subsystem on Eagle is a robust NFS file system intended to provide highly reliable storage for user home directories and NREL-specific software. HFS has a capacity of 182 TB. Snapshots (backup copies) of files in the HFS filesystem are available up to 30 days after change/deletion.

/home

The /home directory on Eagle resides on HFS and is intended to hold small files. These include shell startup files, scripts, source code, executables, and data files.  Each user has a quota of 50 GB.

/nopt

The /nopt directory on Eagle resides on HFS and is where NREL-specific software, module files, licenses, and licensed software is kept.

Parallel File System

The Parallel File System (PFS) on Eagle is a parallel Lustre file system intended for high-performance I/O.  Use PFS storage for running jobs and any other intensive I/O activity. The capacity of 14 PB is provided by 28 Object Storage Servers (OSSs) and 56 Object Storage Targets (OSTs) with 3 Metadata Servers, all connected to Eagle's Infiniband network with 100 Gb/sec EDR. The default stripe count is 1, and the default stripe size is 1 MB.

The PFS hosts the /scratch, /projects, /shared-projects, and /datasets directory.

There are no backups of PFS data.  Users are responsible for ensuring that critical data is copied to Mass Storage or other alternate data storage location.

/scratch

Each user has their own directory in /scratch. Data in /scratch is subject to deletion after 28 days of inactivity.

/projects

Each project/allocation has a directory in /projects intended to host data, configuration, and applications shared by the project.

/shared-projects

Projects may request a shared project directory to host data, configuration, and applications shared by multiple projects/allocations.

/datasets

The /datasets directory on Eagle hosts widely used data sets. 

Common Data Sets

There are multiple big data sets that are commonly used across various projects for computation and analysis on NREL's HPC Systems. We provide a common location on Eagle's scratch filesystem at /datasets, where these data sets are available for global reading by all compute nodes on Eagle. Each data set contains a readme file that covers background, references, explanation of the data structure, and Python examples.

/datasets/NSRDB

The National Solar Radiation Database (NSRDB) is a serially complete collection of meteorological and solar irradiance data sets for the United States and a growing list of international locations for 1998-2017. The NSRDB provides foundational information to support U.S. Department of Energy programs, research, and the general public.

/datasets/WIND

The Wind Integration National Data Set (WIND) Toolkit consists of wind resource data for North America and was produced using the Weather Research and Forecasting Model (WRF).

Node File System

Each Eagle compute node has a local solid-state drive (SSD) for use by compute jobs. They vary in size; 1 TB (standard), 1.6 TB (bigmem), and 25.6 TB (bigscratch), depending on the node feature requested. There are several possible scenarios in which a local disk may make your job run faster. For instance, you may have a job accessing or creating many small (temporary) files, you may have many parallel tasks accessing the same file, or your job may do many random reads/writes or memory mapping.

/tmp/scratch

The local disk is mounted at /tmp/scratch and set under the $LOCAL_SCRATCH environment variable during a job. A node will not have read or write access to any other node's local scratch, only its own. Also, this directory will be cleaned once the job ends. You will need to transfer any files to be saved to another file system. 

For more information about requesting this feature, please see Resource Request Descriptions on the Eagle Batch Jobs page.