Systems

Sapelo2

Sapelo2 is a Linux cluster that runs a 64-bit CentOS 7.5 operating system and it is managed using Foreman and Puppet. Two physical login nodes are available, with Intel Xeon E5-2680 v3 (Haswell) processors and 128GB of RAM and 24 cores per node.

For a subset of compute nodes, internodal communication among them and between these nodes and the storage systems serving the home directories and the scratch directories is provided by a QDR Infiniband network(40Gbps). For another subset of compute nodes, these communications are provided by an EDR Infiniband network.

The cluster is currently comprised of the following resources:

Regular nodes

• 106 compute nodes with AMD Opteron processors (48 cores and 128GB of RAM per node)
• 22 compute nodes with AMD EPYC (Rome) processors (64 cores and 128GB of RAM per node)
• 16 compute nodes with AMD EPYC processors (32 cores and 128GB of RAM per node)
• 42 compute nodes with Intel Xeon Skylake processors (32 cores and 192GB of RAM per node)
• 32 compute nodes with Intel Xeon Broadwell processors (28 cores and 64GB of RAM per node)
• 4 compute nodes with AMD Opteron processors (48 cores and 256GB of RAM per node)

High memory nodes (1TB/node)

• 4 compute nodes with AMD EPYC processors (64 cores and 1TB of RAM per node)
• 4 compute nodes with Intel Xeon Broadwell processors (28 cores and 1TB of RAM per node)
• 1 compute node with AMD Opteron processors (48 cores and 1TB of RAM per node)

High memory nodes (512GB/node)

• 16 compute nodes with AMD EPYC processors (32 cores and 512GB of RAM per node)
• 6 compute nodes with AMD Opteron processors (48 cores and 512GB of RAM per node)

GPU nodes

• 4 compute nodes with Intel Xeon Skylake processors (32 cores and 187GB of RAM) and 1 NVIDIA P100 GPU card per node
• 2 compute nodes with Intel Xeon processors (16 cores and 128GB of RAM) and 8 NVIDIA K40m GPU cards per node
• 4 compute nodes with Intel Xeon processors (12 cores and 96GB of RAM) and 7 NVIDIA K20Xm GPU cards per node

Buy-in nodes

• Various configurations

The queueing system on Sapelo2 is Torque/Moab.

Monitoring Jobs on Sapelo2

---

Back to Top

Slurm Test Cluster (Sap2test)

GACRC is planning to switch the queueing system on Sapelo2 from Torque/Moab to Slurm later this year. At the same time, we will update the cluster OS, from CentOS 7.5 to CentOS 7.8, the compiler toolchains, and the application software packages. Older versions of the applications, currently on Sapelo2, will only be installed in the updated cluster if necessary, upon user request.

In preparation for implementing this major change in the Fall, we are deploying a Slurm development (dev) cluster, that will be available ahead of time. The goal is to give users an environment to modify their workflow scripts to use Slurm and possibly to use newer versions of the applications, prior to the major change. All job submission scripts will need to be changed, because Slurm uses different syntax from Torque/Moab, as summarized in Migrating from Torque to Slurm. We strongly encourage everyone to fully test their ported workflow scripts on the Slurm dev cluster, to ensure a smooth transition to the new system later in the year.

This dev cluster is intended to allow users to port their workflow scripts to Slurm, and it is not a platform for users to run jobs extensively. This dev cluster currently has the following resources:

Regular nodes

• 40 compute nodes with AMD Opteron processors (48 cores, 128GB RAM per node)
• 24 compute nodes with AMD EPYC processors (64 cores, 128GB RAM per node)
• 6 compute nodes with AMD EPYC processors (32 cores, 128GB RAM per node)
• 4 compute nodes with AMD Opteron processors (48 cores, 256GB RAM per node)
• 1 compute node with Intel Broadwell processors (28 cores, 64GB RAM per node)
• 1 compute node with Intel Skylake processors (32 cores, 192GB RAM per node)

High memory nodes (512GB)

• 2 compute nodes with AMD EPYC processors (32 cores, 512GB RAM per node)
• 4 compute nodes with AMD Opteron processors (48 cores, 512GB RAM per node)

GPU node

• 1 compute node with Intel Skylake processors (32 cores, 192GB RAM per node) and a P100 GPU card

Storage

The user's home directory (/home), scratch directory (/scratch), and each group's work directory (/work) on the Slurm test cluster are the same file systems as on Sapelo2. So there is no need to transfer data between Sapelo2 and Slurm test cluster. If you have Sapelo2 specific settings in your dotfiles (for example in .bashrc or in software specific configuration files), those might need to get changed when you work on Sap2test. The environment variable GACRC_CLUSTER stores the test cluster name, and can be used to set up a cluster specific dotfile to use on the test cluster.

However, Sapelo2's /usr/local file system and therefore the applications installed on Sapelo2 are not available on the Slurm test cluster.

Training material

To help users familiarize with Slurm and the test cluster environment, we have prepared some training videos that are available from the GACRC's Kaltura channel at https://kaltura.uga.edu/channel/GACRC/176125031 (login with MyID and password is required). Training sessions and slides are available at https://wiki.gacrc.uga.edu/wiki/Training

Getting Help

If you run into any issues on the test cluster or have any questions or suggestions, please let me know via the online form below, as it will reach all the GACRC staff members:

Support for Slurm test cluster

Migrating from Torque to Slurm

---

Back to Top

Teaching cluster

The teaching cluster is a Linux cluster that runs a 64-bit Linux, with Centos 7.8. The login node is a VM that has 4 cores (Intel Xeon Gold 6230 processor) and 16GB of RAM. An Ethernet network (1Gbps) provides internodal communication among compute nodes, and between the compute nodes and the storage systems serving the home directories and the work directories.

The cluster is currently comprised of the following resources:

• 30 compute nodes with Intel Xeon X5650 2.67GHz processors (12 cores and 48GB of RAM per node)
• 2 compute nodes with Intel Xeon L7555 1.87GHz processors (32 cores and 512GB of RAM per node)
• 4 NVIDIA Tesla (Kepler) K20Xm GPU cards. These cards are installed on one host that has dual 6-core Intel Xeon CPUs and 48GB of RAM

The queueing system on the teaching cluster is Slurm.

Software Installed on the teaching cluster

The list of installed application is available at Software page.