The Open Science Computing Ecosystem at the Texas Advanced Computing Center (TACC) Siva Kulasekaran and Doug James November 13, 2014 siva@tacc.utexas.edu, djames@tacc.utexas.edu
The TACC Ecosystem Stampede Maverick HPC Jobs Vis & Analysis 6400+ Nodes Interactive 10 PFlops Access 14+ PB Storage 132 K40 GPUs Wrangler Lonestar Data Intensive HTC Jobs Computations 1800+ Nodes 10 PB Storage 22000+ Cores Stockyard High IOPS 146 GB/node Shared Workspace 20 PB Storage 1 TB per user Corral Rodeo Project Workspace Data Collections Cloud Services 6 PB Storage User VMs Databases XXX VCores IRODS XXX PB Vis Lab Ranch Immersive Vis Tape Archive Colaborative 160 PB Tape Touch Screen 1TB Access 3D Cache
Wrangler Hardware @TACC Three primary subsystems: – A 10PB, replicated Mass Storage Subsystem 10 PB disk storage system. (Replicated) – An embedded IB Interconnect 120 Lanes (56 Gb/s) non-blocking analytics capability of Access & several thousand Analysis System 96 Nodes cores. 128 GB+ Memory Haswell CPUs – A high speed global Interconnect with object store 1 TB/s throughput High Speed Storage System • 1TB/s 500+ TB 1 TB/s • 250M+ IOPS 250M+ IOPS
Stampede - High Level Overview • Base Cluster (Dell/Intel/Mellanox): – Intel Sandy Bridge processors – Dell dual-socket nodes w/32GB RAM (2GB/core) – 6,400 nodes (102,400 cores, 2.2PF Peak) – 56 Gb/s Mellanox FDR InfiniBand interconnect • Co-Processors: – 6,880 Intel Xeon Phi Coprocessors – First large scale production Phi system – 7.4+ PF peak performance • Max Total Concurrency: – exceeds 500,000 cores – 1.8M threads • Entered production operations on January 7, 2013
TACC Services: Data • Online, long term, high integrity file or relational – Corral • Online short term, high capacity (PBs), high speed – Stockyard, /scratch • Offline, long term, ultra high capacity (tape): – Ranch • Object store: – Wrangler (coming soon) • Data management and curation, Data analysis and statistics • Static and dynamic web services, Database applications
Storage Solutions at TACC Supported Services § Storage of input data, results, processed data, interim data products § Data management services to allow for controlled sharing of data with colleagues, collaborators, and public § Databases to store and query structured data § GIS extensions for geographically structured data § Web services to integrate data with other portals and data gateways § Capability to develop storage solutions for PHI data Not Supported Services § Systems backup and restoration services § Administrative data storage
TACC Services: Computing • Batch – Stampede, Lonestar, Wrangler • Interactive – Maverick • Short term, ephemeral virtual machines: – Rodeo (and soon-to-be-announced systems) • Persistent VM (for datasets hosted at TACC) – Available: contact us • Hadoop: – Rustler, Wrangler • Computing research platforms: – Chameleon and others
TACC Services: Other • Data visualization • Ticket support • Training • Web portal/gateway development and hosting • HPC collaboration and consulting (optimization, parallelization, workflow support, porting, etc.) • Proposal support
TACC User Portal Easy to get started! portal.tacc.utexas.edu
Siva Kulasekaran and Doug James siva@tacc.utexas.edu djames@tacc.utexas.edu For more information: www.tacc.utexas.edu
Recommend
More recommend
