Quick Menu
Applications howtos
Archived pages
Support
- Phone 3004
Sidebar
start
This is an old revision of the document!
Table of Contents
HPC Cluster @ DCMC
The Cluster is made of some servers and blades organized in 3 racks physically located in the Cluster Room and Datacenter Room in Mancinelli site. The main hardware components where part of the former Labs cluster, new acquisitions and one blade from Mathematics Dept. Datacenter donated from a Research group of DICA.
Hardware
| Server or Blade | # nodes and processors | # cores/processor | Ram (GB) | Local storage | Network interfaces |
|---|---|---|---|---|---|
| Gandalf | 6 nodes Dell with 2 Intel Xeon | 4 | 8 | 1 HD SAS 73GB | 1Gbit Ethernet for Management and Data |
| Merlino | 9 nodes Dell with 2 AMD Opteron | 4 | 16 | 1 HD SATA 80GB | 1Gbit Ethernet for Management, 10Gbit Ethernet for Data |
| Blade Morgana | 11 nodes Dell with 2 Intel Xeon | 4 | 24 | 2 HD SAS 146GB Raid0 | 1Gbit Ethernet for Management, 20Gbit Mellanox Infiniband for MPI, 10Gbit Ethernet for Data |
| Covenant | 1 node Hp with 4 Intel Xeon | 10 | 256 | 2 HD SAS 1TB Raid0 | 1Gbit Ethernet for Management, 10Gbit Ethernet for Data |
| Masternode | 1 node Dell with 2 Intel Xeon | 4 | 24 | 2 HD SAS 1TB Raid1 for OS, 4 HD SAS 2TB Raid5 for shared Scratch, 35TB Storage for Home | 1Gbit Ethernet for Management, 1Gbit Ethernet for Frontend/login, 1Gbit Ethernet for nodes console (iDRAC, ILO), 2 10Gbit Ethernet for Data, Fiber Channel 8Gbps for Storage, 1 10Gbit Mellanox Infiniband for Infiniband control |
The Masternode provides:
- A file system /opt/ohpc/pub/ on local Raid5 storage (up to 2 TB) for applications and process management, shared via NFS on Gigabit network
- A file system /homes on Dept. high performance Storage (up to 5 TB) for homes and user data shared via NFS on Gigabit network (Morgana nodes have 1 Gb bandwidth guaranteed due to 10Gb Ethernet hardware)
- Login for users
- Node provisioning software (installation of new nodes and node rebuild in less than 5 min.)
- PBSPro master for scheduling and resource management
- Support for Singularity container technology
- Infiniband network controller (via software)
- Application software for compute nodes
- Monitoring and utility software (Ganglia, Nagios, Hpc website and web interface to PBSPro submission)
Software
The cluster has been installed following the directions of the OpenHPC Project, an open source project backed by Intel and many software and hardware HPC player and supported by a strong developer community. The main reason for this choice is that they committed to support the upgrade of the software platform following the evolution of the single components (OS, compilers, MPI distributions, toolchains components, utility software, hardware support, container technology) and also for their independence from a specific vendor or technology. The installed software is divided in Management and operations software and Application software.
Management and operations software
- OpenHPC 1.3.1 (July 2017) for Linux CentOS 7.3 nodes image
- PBS Pro 14.1 (Altair released PBSPro as open source in mid 2016)
- Compilers: GCC 7.1.0, Intel C++/Fortran 17.0.4 ver. 2017 update 4
- MPI: OpenMPI 1.10.7, Intel MPI 2017.3, MPICH 3.2, MVAPICH2 2.2
- Ganglia for monitoring cluster performance (public access)
- Nagios for monitoring nodes health status (no public access)
Application software
- Abaqus 6.14
- Abaqus 2017
- Ansys 18.2.2 (Fluent and LSDYNA)
- Comsol 5.3
- Matlab R2017b with parallel support
Usage guides
Applications specific instructions
start.1554477213.txt.gz · Last modified: 2019/04/05 17:13 by druido
