Date post: | 19-Jan-2016 |
Category: |
Documents |
Upload: | june-chase |
View: | 217 times |
Download: | 0 times |
Sharon Broude GevaDirector of Advanced Research Computing (ARC)University of [email protected]://arc.umich.edu
NewAdvanced Research
Computing Projects & Services at U-M
2015 CASC Fall Meeting - 10/15/2015
Advanced Research Computing
U-M Data Science Initiative
● U-M investment of $100M over next 5 years○ Hire 35 new faculty over the next four years and
engage existing faculty across campus;○ Support interdisciplinary data-related research
initiatives and foster new methodological approaches to big data;
○ provide new educational opportunities for students pursuing careers in data science;
○ expand U-M’s research computing capacity; and○ strengthen data management, storage,
analytics, and training resources.● 1500+ registrants for kick-off symposium last week
U-M Data Science Initiative
Faculty affiliates, challenge grants, graduate certificate, industry engagement
Data Science infrastructure build and operations
Data Science consultants and training
CC*DNI Award: MI-OSiRIS
● PI: Shawn McKee (U-M Physics, ARC)● Co-PI’s: Swany (IU), Gossman (WSU), Merz
(MSU)● $4.9 Million (NSF)● Will provide a distributed, multi-institutional storage
IF that allows researchers at any of the three campuses to read, write, manage, and share data from their computing facility locations
● Goal - Provide transparent high-performance access to the same storage IF from well-connected locations on any of the three campuses
CC*DNI Award: MI-OSiRIS
● Will include network discovery, monitoring and management tools and creative use of CEPH features
● Users get customized data interfaces for their multi-institutional data needs
● Seamless rebalancing and expansion of storage● Data sharing, archiving, security, and life-cycle
management implemented and maintained with single distributed service
● Data IF view for each research domain can be optimized
CC*DNI Award: MI-OSiRIS
Center for Network and Storage-Enabled Collaborative Computational Science
Build and operations at U-M
MRI Award: ConFlux
● PI: Karthik Duraisamy (Aerospace Engineering)● $3.5 Million ($2.42M NSF + $1.04M cost-share) ● Designed to enable HPC simulations to interface
with large datasets while running● To refine complex physics-based models with Big
Data techniques (For cardiovascular disease; turbulence; clouds, rainfall, climate; dark matter, dark energy; material property prediction)
● CPU’s + GPU’s, large memory, ultra-fast interconnect, 3 PB hard drive
● Optimized hardware for machine learning● Plan to expand availability to researchers and
schools outside the grant team
MRI Award: ConFlux
Center for Data-Driven Computational Physics
Technical design, build, and operations
Turbo: High Performance Research Storage
● Isilon scalable storage for U-M researchers● Configured to be easily shareable with on-campus
resources such as the Flux HPC cluster, as well as off-campus systems and collaborators
● Performance sufficient for both IO-intensive operations and bulk file access, allowing researchers to work with data in place and avoid excessive data staging
● Primary tiered storage - Hybrid w/SSDs● Tuned for large files (1MB or greater) but capable of
handling small files (e.g., documents, spreadsheets, etc.)
Turbo: High Performance Research Storage
● NFSv3 and NFSv4 access (for Linux, Mavericks, Yosemite, Windows 7+)
● NFSv4 w/Kerberos access (Linux and Mavericks)● Two security levels:
○ regulated and/or sensitive data (PHI only w/Kerberos)
○ non-sensitive data ● Globus available for volumes that do not contain
PHI, for sharing and hosting data for external collaborators and institutes.
● Currently 1PB usable, replicated to 1PB DR● Cost: $19.20 / TB / Month (Optional daily snapshots
available at no cost)
Turbo: High Performance Research Storage
Technical design, build, and operations