Bridges: Connecting Researchers, Data, and HPC

Presenter(s): Nick Nystrom (PSC)

Principal Investigator(s): Nick Nystrom (PSC)

Presentation Slides

Bridges is a new kind of supercomputer being built at the Pittsburgh Supercomputing Center (PSC) to empower new research communities, bring desktop convenience to supercomputing, expand campus access, and help researchers facing challenges in Big Data to work more intuitively. Funded by a $9.65M NSF award, Bridges consists of tiered, large-shared-memory resources with nodes having 12TB, 3TB, and 128GB each, dedicated nodes for database, web, and data transfer, high-performance shared and distributed data storage, the Spark/Hadoop ecosystem, and powerful new CPUs and GPUs. Bridges is the first production deployments of Intel's new Omni-Path Architecture (OPA) Fabric, which will interconnect its nodes and storage. Bridges emphasizes usability, flexibility, and interactivity. Widely-used languages and frameworks such as Java, Python, R, MATLAB, Hadoop, and Spark benefit transparently from large memory and the high-performance OPA fabric. Virtualization enable hosting web services, NoSQL databases, and application-specific environments and enhances reproducibility. Bridges, allocated through XSEDE, is available at no charge to the open research community. Bridges is also available to industry through PSC's corporate programs.

Design of Experiments and Big Data Analytics for Energy Efficient Buildings

Presenter(s): Pragnesh Patel (NICS)

Principal Investigator(s): Joshua New (ORNL)

Presentation Slides

A central challenge in the domain of energy efficiency is being able to realistically model a specific class of building and scaling those classes up to the entire United States building stock across ASHRAE climate zones, then projecting how specific retrofits or retrofit packages would maximize return-on-investment for subsidies through federal, state, local, and utility tax incentives, rebates, and loan programs. Nearly all projections regarding energy savings, for any of the plethora of technologies required to address the need for US energy security, are reliant upon accurate models as the central primitive by which to integrate the national impact with meaningful measures of uncertainty, error, variance, and risk. This challenge is compounded by the fact that buildings, unlike cars or planes, are manufactured in the field at the time of construction based on one-off designs with a median lifespan of 73 years. Due to variance of building materials, construction, and equipment (and the necessary flux of these over time), a given building is unlikely to closely resemble the prototypical building class. Therefore, each building needs to be modeled individually and precisely to achieve optimal retrofit and construction practices. We have developed design of experiement for calibrating building energy models, which minimize the number of simulations required while maximizing the statistical resolution of analysis results. Initial statistical analysis of parametric ensembles using techniques such as multiple analysis of variance (MANOVA) and a software infrastructure tying together several machine learning packages (MLSuite) have recently pushed the cutting edge of building energy analysis from about 10 inputs and 12-24 outputs to156 inputs and 96 outputs. The science-enabling software infrastructure has been improved as part of this project include improving R code for design of experiments along with R analysis code while quickly instantiating R on every parallel node/core, integration of Energyplus code for large-scale simulation runs with OpenDIEL workflow system along with pre and post processing data analysis codes.