Final Report: A Transitional Computational Platform to Migrate Parachute Simulation from Workstation to HPC
Abstract
Supported by the DURIP grant W911NF-15-1-0403 for 'A Transitional Computational Platform to Migrate Parachute Simulation from Workstation to HPC', we purchased a parallel cluster from the Advanced Cluster Technologies, Inc.. This parallel cluster is named 'Intruder', after a type of sports parachute. It consists of one head node, 21 computing nodes (20 CPU (Central Processing Unit) nodes and 1 GPU (Graphic Processing Unit) node), connected with 56Gb/s InfiniBand and 1000MB/s Ethernet. Each node was populated with dual Eight-Core Intel E5-2630v3 'Haswell' 2.4GHz processors with different size of RAM (Random-access Memory) and Storage. The head node and compute node have 32GBof RAM for each, and the GPU node has 128GB of RAM. A 32TB of network file system using RAID6 (Redundant Array of Independent Disks) is installed in the head node, and is shared with other nodes. Each node also has a clone of the operation system on its local disk: 2TB SSD on head node, 1TB SATA drive on compute nodes, and 240GB SSD on GPU node. The parallel computing can be further accelerated by including the GPU node, which contains seven NVIDIA Tesla K40 GPUs with 12GB of RAM for each device.
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 2016
- Accession Number
- AD1064258
Entities
People
- Xiaolin Li
Organizations
- Stony Brook University