150 likes | 165 Views
Navigator Desde 2015 e m reinstalação (Lustre, CentOS , bibliotecas, compiladores) 24 cores por nó 96 GB por nó 164 nós 3936 cores 15,7 TB FDR infiniband 2:1 Lustre / home +/ scratch 220 TB. Em fase final de concurso Cluster 4 0 cores por nó high-memory nodes com 384 GB
E N D
Navigator Desde 2015 em reinstalação (Lustre, CentOS, bibliotecas, compiladores) 24 cores por nó 96 GB por nó 164 nós 3936 cores 15,7 TB FDR infiniband 2:1 Lustre /home +/scratch 220 TB Jornadas de computação científica
Em fase final de concurso Cluster 40 cores por nó high-memory nodes com 384 GB 4 nós com 2 GPUs V100 NVIDIA 1 nó SMP (4 sockets) com 3 TB de memória Discos locais SSD Infiniband EDR (100 Gb/s) non-blocking Sistema de armazenamento de alto desempenho - Lustre 1,2 PB úteis expansíveis 18 GB/s IOR MetaData Servers redundantes 8 interfaces externos EDR (4 OSS+ 2x2 MDS) Software de gestão • Aplicações • pós-processamento • Aplicações especiais (ISV…) • Metagenómica e Genómica • Dinâmica molecular e aplicações que possam ser aceleradas por GPUs • Processamento direto de dados • Novos workflows– uso de containers • AI (testes– Tensor Flow incluído nos benchmarks) Jornadas de computação científica
“Tape library” para backups persistentes Recursos humanos e formação Recrutamento e formação de recursos humanos qualificados Usar o conhecimento dos atores internacionais relevantes: PRACE, vendedores quando apropriado ERASMUS+ Integração do Navigator na rede PRACE Colaboração com outras infraestruturas do Roteiro Nacional BIN GenomePT Engage/SKA (projeto AENEAS) Colaborações com empresas (Digital Hub da região Centro) Projeto piloto a começar brevemente Jornadas de computação científica
Laboratory for Advanced Computing (LCA) Mission • Supply HPC services to scientists and companies • Generic and specialized supercomputing services • Post-processing, storage • Industry 4.0 • Training and dissemination • Support advanced computer courses (libraries/languages and algorithms) • Parallelization and HPC open source software workshops • Dissemination and collaboration in promoting of HPC – ex. http://supercomputer.pt Jornadas de computação científica
Activities • 2007-2014 (Mlipeia) • 7 nacional calls for CPU time - 220 projects, 20 M core-hours. PI’sfrom 8 Universities, 3 AssociatedLaboratories, 1 StateLaboratory. • Material Science, QCD, CFD, Molecular Dynamics, Astrophysics, Cosmology,etc. • Publicaccess rules • Users training • HPC workshops • 2015 - (Navigator) • 26 M core-hours • New: firepropagationsimulations (report submitted by ADAI to the government) Jornadas de computação científica
International connections • PRACE (Partnership for Advanced Computing in Europe) • founding member • Participation in PRACE preparatory and implementation european projects 1IP-5IP • Partners: IST e Univ. Evora • RISC - A Network for Supporting the Coordination of Supercomputing Research between Europe and Latin America • Member of IDC HPC (now Hyperion) Technical Computing Advisory PaneI MoU – Berlin 2007 PRACE AISBL – Brussels 2010 Jornadas de computação científica
PRACE Hosting Members offering of core hours on7 world-class machines JUQUEEN: IBM BlueGene/Q GAUSS/FZJ Jülich, Germany SuperMUC: IBM GAUSS/LRZ Garching, Germany NEW ENTRY 2016 MareNostrum: IBMBSC, Barcelona, Spain Hazel Hen: Cray GAUSS/HLRS, Stuttgart, Germany Piz Daint: Cray XC 30 CSCS Lugano, Switzerland MARCONI: Lenovo CINECA Bologna, Italy CURIE: Bull Bullx GENCI/CEA Bruyères-le-Châtel, France
International connections • Navigator to beinserted in Tier-1 PRACE network • 164 computing nodes • 2x Xeon E5-2697v2 -> 24 cores/node ( 3936 total) • 96 GB/node • Interconnectinfiniband FDR 56 Gbit/s • 180 TB Lustre central storage Tier-0 European centres Tier-1 National centres Tier-2 Regional/Universitycentres Jornadas de computação científica
E-infrastructure for HPC • 2014 • Type 1 scientific infrastructure included in the national roadmap of FCT • 2017-2020 • Financing • Collaborations with other infrastructures • INCD and RCTS • BIN / Viravector • ENgAGE SKA • GenomePortugal … • Research Centers and entreprises Jornadas de computação científica
LCA development • Equipment • Central storage (> 500 TB) with several storage tiers • Supplementary cluster with GPU nodes and large memory nodes (post-processing, genome sequencing, GPU accelerated processing) • Possibly deep learning hardware for several applications (i.e. medical imaging) • Human resources • Training Jornadas de computação científica
PRACE and SKA • European HPC ecosystem developments • PRACE 2 • Centers of Excelence • European Data infrastructure (EDI) / EuroHPC for pre-exascale and exascaleeuropean systems Jornadas de computação científica
PRACE and SKA • The most recent PRACE document for EDI mentions SKA as a very important HPC use case • data-intensive (not much floating point arithmetic) • memory and I/O bandwidth requirements are enormous • complex job scheduling • Continuous operation, need data buffer • Exascale processing power needed – which hardware? Jornadas de computação científica
LCA and SKA in Portugal Hardware • Colaboration with University of Evoraregarding the aquisition of a new cluster • Colaborationwith ENgAGESKA for new investments in HPC hardware (computing and storage) System management • Collaboration for • job scheduling • Data processing workflows Training ? Jornadas de computação científica