Supporting the UH Research Mission with HPC and Data Services Ron Merrill merrill@hawaii.edu
What is HPC? High Performance Computing (HPC) Scales across many nodes, needs a low latency network High Throughput Computing (HTC) Scale out, each node independent, same code different data Advanced Computing Bigger than your laptop Includes HPC, HTC, cloud, storage, vizualization, etc
Advanced Computing at UH ITS We use a condo model for campus computing Base system – FREE to all UH students, staff and affiliates PI’s can purchase nodes or time for priority access Condo model Common to many US research universities Alternative: Departmental and lab scale resources Also common and co-existing here at UH
Condo-nomics Sharing economy: most like Air B&B Increase efficiency by increasing asset utilization Expand access to goods Provide income to owners Enabled by low transaction costs Air B&B enabled by the web Condo model enabled by job schedulers, ex. SLURM The owner shared nodes are like an Air B&B, but one where you stay for free. Then the bears come home and quickly kick you to the curb. The income to owners is free scratch space, free sysadmin , free procurement, free hosting. Kill queue users that have check pointing or short enough jobs end up making progress.
Cray CS-300 10/2014 – 4/14/2015 $1.79M Delivered, accepted, early adopters 4/15/2015 Open to all UH students, staff and affiliates Hardware – 3800 cores / 28.9 TB RAM 178 nodes, 20 core/node, 128GB RAM 6 nodes, 40 core/node, 1024 TB 582 TB Lustre filesystem
Cray CS-300
March 2016 HPC Upgrade 8 PI’s, $712k Hardware 33 nodes, 20 core/256GB RAM 58 nodes, 24 core/128GB RAM 1 GPU node, 20 core/128GB RAM, 2xNVidia K40 Core count: 3800 -> 5872, 54%
Cray CS-300
Rear View Layout White are all of the original standard nodes.
HPC
Fall 2016 HPC Upgrade 4 PI’s, $155k Hardware Core count: 8 nodes, 24 core/256GB RAM 1 large memory node, 72 core/1024GB RAM 2 nodes, 24 core/128GB RAM 1 GPU node, 20 core/128GB RAM, 2xNVidia K40 Core count: 3800 -> 5872 (54%) -> 6204 (63%) -> ~6828 (80%)
So far… 280 UH researchers, faculty and students attended on-boarding training received accounts UH ITS HPC cluster has delivered... 39 million CPU hours 783,000 compute jobs
Research Data Services Data Management Services Enable workflows by integrating storage and compute assets Relational and NoSQL database design Sharing and distribution Data Management Plans Federal funding requirement Science Gateways Domain focused data repositories
Storage: Data Services Foundation Value Storage NetApp NAS filer – available only to UH-HPC users for a fee CIFS deployment on hold OwnCloud “an open source, self-hosted file sync and share app platform” 10 years old Deployment: EPSCOR and other early adopters
CI People Gwen Jacobs, Director of Cyberinfrastructure (CI) Sean Cleveland, CI Research Scientist David Schanzenbach, Lead Software Architect Michelle Choe, EPSCoR Program Assistant TBD, CI Software Engineer
Thank you!