Information Technology Services (ITS) HPC Infrastructure HPC Day October 15 th 2014 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC@ODU Introduction High Performance Computing Resources. Storage Software Support staff Education and Outreach 2 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC@ODU Introduction Supporting research computing with parallel compute environments using MPI and OpenMP protocols. Pre 2013, ZORKA & NIKOLA were providing computation resources to the campus researchers. Some examples of resources in these clusters: 40 Dell PE 1950 nodes, 4 cores each & 8GB RAM. 4 Dell PE R900 nodes, 16 cores each & 32GB RAM. 7 Sun Fire X46000 M2 Nodes, 32 cores each & 64 GB RAM. 17 APPRO nodes , Tesla M2090 GPU’s, 12 cores & 48GB RAM . NFS based scratch space mounted on all compute nodes (about 2TB). Research Mass storage, accessible from Head Node (approximately 60TB of data plus tape storage). 3 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC Turing Cluster Base Configuration New HPC cluster , primarily deployed to support GROMACS. Funding support from Dr. Vernier, Bio-Electric research center & ITS. Initial specification for Turing Cluster is as follows: FDR based infiniband switches. Dell C8000 chassis with eight (8) SLEDS Each SLED has 16 cores (E5-2660) & 32GB RAM. (1) Head Nodes , Dell R720 server with 128GB memory 1Gbps switching hardware with 10Gbps uplinks Base for expansion of research computing infrastructure at ODU. Available to Researchers for computation since Fall 2013. 4 Information Technology Services | www.odu.edu/its | itshelp@odu.edu 4
HPC Turing Cluster Expansion Integration of 8 computes nodes purchased by Dr. Gangfeng Ma (Civil & Environmental Engineering Department) Thirty Six (36) compute SLED’s added by ITS 720 compute cores (E5-2660v2, ivy-bridge) 128GB per node, total 24TB. 2x500GB local disk per compute node. Additional Head Node for redundancy ( Dell R720 server with 128GB memory) Separate Login Node for optimal compute environment. 5 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
Turing Cluster Summer 2014 upgrade Seventy Six(76) CRAY CS-GB-512X compute nodes 1520 compute cores (E5-26670v2, ivy-bridge) 128GB per node. 2x250GB local disk per compute node. Four (4) High Memory Nodes 32 compute cores (E5-4610v2, ivy-bridge) per node 768 GB per node. Ten(10) Xeon Phi Nodes Each node has 2 Xeon Phi (60 core) co processors. Each node has 20 cores (E5-2670v2, ivy-bridge) , 128 GB memory Infiniband (FDR based) compute backbone upgraded 324 level 1 FDR interfaces. 144 level 2 backbone interfaces. 6 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
Turing Cluster Total Total Total Compute GPU/MIC Cores/ GPU Co- GPU/MIC Storage Total Storage Quantity Cores Cores Item Description Sockets Socket Processors cores/Socket Mem(GB) (TB) Mem(GB) (TB) PowerEdge C8220, 2 x E5- Dell C8000 chassis 2660(2.2GHz), Mellanox single port 8 128 0 Sandy-Bridge nodes FDR 2 8 0 0 128 1 1024 8 PowerEdge C8220, 2 x E5- Dell C8000 chassis 2660v2(2.2GHz), Mellanox single port 36 720 0 Ivy-Bridge nodes FDR 2 10 0 0 128 1 4608 36 PowerEdge C6220, 2 x E5- Dell C6000 chassis 2660(2.2GHz), Mellanox single port 8 128 0 Sandy-Bridge nodes FDR 2 8 0 0 128 1 1024 8 PowerEdge C6220, 2 x E5- Dell C6000 chassis 2660(2.2GHz), Mellanox single port 12 192 0 Sandy-Bridge nodes FDR 2 8 0 0 128 1 1536 12 CRAY SR5110 CRAY CS-GB-512X, 2 x E5- chassis Ivy-Bridge 2670v2(2.5GHz), Mellanox single port 76 1520 0 nodes FDR 2 10 0 0 128 0.5 9728 38 CRAY SR5110 CRAY CS-GB-512X, 2 x E5- chassis Ivy-Bridge 2670v2(2.5GHz), 2 x Xeon Phi 5110P, 10 200 1200 nodes Mellanox single port FDR 2 10 2 60 128 1 1280 10 Intel R2304LH2HKC, 4 x E5-4610v2, 4 128 0 High Memory Nodes Mellanox single port FDR 4 8 0 0 768 4 3072 16 APPRO 1426G4, 2 x Intel Xeon X5650 CRAY Appro GPU (2.67Ghz), 4 x NVidia M2090 GPU 17 204 34816 Nodes processor, Mellanox single port QD 2 6 4 512 48 1 816 17 7 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC Turing 8 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
STORAGE HPC Legacy Clusters had NFS based scratch space mounted on all compute nodes. Research Mass storage, accessible from Head Node (approximately 60TB of disk plus tape storage). Complete redesign of the computational storage infrastructure. EMC Isilon based scale out NAS storage . Integration of additional 430TB for computational research. Integration of 36TB of LUSTRE based scratch space. 9 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
Turing Cluster Queuing Strategy Current Job Scheduler is SGE. Four separate job queues Traditional computational resources High memory nodes APPRO GPU nodes Nodes with dual Intel Xeon Phi Fair Queuing Strategy Shared computational resources Compensates user over time 10 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC Software Packages Some Software Packages on HPC Turing Cluster: COMSOL (finite element analysis & simulation software for physics and engineering) MATLAB (Numerical computational software, plotting of functions, implementation of algorithms) CHARMM (Molecular dynamics simulation & analysis software) METIS (Software package for graph partitioning) GAUSSIAN (Software package for computational biology) GROMACS (Molecular dynamics simulation & analysis software for proteins, lipids and nucleic acids) R (Software programming language & software environment for statistical computing and graphs) CLC Bio (Software package for analysis of biological data) MOLPRO (Software package used for quantum chemistry calculations) DDSCAT (Software package for calculating scattering and absorption of irregular particles) 11 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
Globus Globus (SaaS) 12 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC@ODU Support Structure Early 2013 , HPC support staff challenges Two engineers moved on to other opportunities One systems administrator position realigned for HPC support. Started recruitment process to add two position focused on HPC Addition to two (2) dedicated HPC systems engineers. Jeaime Powell Terry Stilwell A computational scientist position is ready to to be recruited shortly. 13 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
Education & Outreach Effective means of communicating with researchers and students on campus regarding available resources and services Quarterly HPC newsletter HPC day. New faculty outreach . HPC Advisory Committee. 14 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
HPC DAY THANK YOU 15 Information Technology Services | www.odu.edu/its | itshelp@odu.edu
Recommend
More recommend