Cornell Leads New National Science Foundation Federated Cloud Project
Initial users of the cloud federation were selected based on the diversity of their data analysis requirements and cloud usage modalities.
Cornell University will lead a five-year, $5 million project sponsored by the National Science Foundation (NSF) to build a federated cloud comprised of data infrastructure building blocks (DIBBs) designed to support scientists and engineers requiring flexible workflows and analysis tools for large-scale data sets, known as the Aristotle Cloud Federation.
The federated cloud will be deployed at Cornell University (CU), the University at Buffalo (UB), and the University of California, Santa Barbara (UCSB) and shared by seven science teams with over forty global collaborators.
David Lifka, director of the Cornell University Center for Advanced Computing (CAC) will lead the project with colleagues Tom Furlani, director of the UB Center for Computational Research, and Rich Wolski, professor of Computer Science at UCSB.
Initial users of the cloud federation — earth and atmospheric sciences, finance, chemistry, astronomy, civil engineering, genomics, and food science — were selected based on the diversity of their data analysis requirements and cloud usage modalities. Their use cases will demonstrate the value of sharing resources and data across institutional boundaries. The overarching goal is optimizing “time to science”— the actual time it takes a researcher to obtain their scientific results. The elasticity provided by sharing resources means researchers don’t have to wait for local resources to become available to get their science started.
Metrics provided by UB's XDMoD (XD Metrics on Demand) and UCSB's QBETS (Queue Bounds Estimation Time Series) will enable researchers and administrators to make informed decisions about when to use federated resources outside their institutions.
“Cloud-based systems are rapidly becoming a key component in the support of research programs in academe and industry. By adding cloud metrics to XDMoD, researchers and senior leaders will be able to obtain detailed operational metrics of cloud systems in order to improve the efficiency of jobs run on the cloud, as well as measure overall cloud performance,” said Furlani. “Efficient use of federated clouds requires the ability to make predictions about where a workload will run best,” added Wolski. “Using XDMoD data and cloud-embedded performance monitors, QBETS will make it possible to predict the effects of federated work-sharing policies on user experience, both in the DIBBs cloud and in the Amazon Web Services (AWS) Cloud.”
"The goal of the Aristotle Cloud Federation is to develop a federated cloud model that encourages and rewards institutions for sharing large-scale data analysis resources that can be expanded internally with common, incremental building blocks and externally through meaningful collaborations with other institutions, commercial clouds, and NSF cloud resources," said project PI Lifka. The project name—Aristotle—was chosen because Aristotle’s concept “the whole is greater than the sum of its parts” reflects the multi-institutional synergy and collaborations that the federation aspires to create.
The project will implement a new allocations and accounting model that will allow institutional administrators to track utilization across federated sites and use this data as an exchange mechanism between partner sites. This data will demonstrate the potential benefits of sharing institutional resources such as deploying local infrastructure that is right-sized for steady state usage rather than irregular peak loads.
Federation components, documentation, and best practices developed in this grant will be provided to the national community with the information necessary to create customized Virtual Machine instances, leverage resources at federated sites, burst to AWS, access, move, and share large-scale data, and deploy new cloud federations.
Cloud provider AWS will collaborate with the federation developers and scientists. “We are excited to work with the Aristotle team to provide cost-effective and scalable infrastructure that helps accelerate the time to science," said Jamie Kinney, senior manager Scientific Computing, Amazon Web Services, Inc.
Scientists will use the federation to solve data challenges. “We plan to use Aristotle to exploit cloud-based parallelism and perform asynchronous, interactive analysis of complex environmental models that generate thousands of data files" said Patrick Reed, a Cornell University Civil and Environmental Engineering researcher who collaborates with University of North Carolina, Chapel Hill and Penn State engineers. “We will use Aristotle to enhance our decision management tools so that we can solve problems of increasing complexity such as helping cities to better manage their drought risks.”
According to Varun Chandola, a computer science and engineering researcher at the UB, massive troves of geospatial data such as earth observation and climate simulations are scattered around the world within the data archives of researchers, government, and the private sector. Chandola is working with colleagues at NASA Ames, Oak Ridge National Laboratory, and several universities on streamlining the integrated visualization and analysis of geo-data. “We plan to use Aristotle to develop a cloud-based solution that allows researchers to seamlessly integrate heterogeneous geo-data from a variety of sources into a cloud-based analysis engine," Chandola said.
“Research scientists and their collaborators are gathering sensor data and scientific images to optimize food productivity and security,” said Kate McCurdy, director of the Sedgwick Reserve, a 5,896 acre nature reserve in California. “The scientists wish to combine this data with images taken by the general public and stored in commercial clouds,” she explained. “By combining campus clouds and commercial cloud services, the federated cloud approach implemented by Aristotle will provide the data structure we need.”
“This award continues NSF’s multi-year strategy to stimulate exploration of scalable and sustainable data infrastructure models that facilitate collaborative research across disciplines and institutions,” said Amy Walton, program director, Advanced Cyberinfrastructure Division, NSF. “By experimenting with cloud usage metrics, collaborating with a commercial cloud vendor, and exploring pricing/trading allocation mechanisms, the project will provide valuable information about how the innovations work in a range of situations, and how this ‘market approach’ integrates within the larger research ecosystem.”
“Sharing cloud computing and storage assets between institutions and bursting to commercial clouds when appropriate is definitely a model worth a serious trial,” said Robert A. Buhrman, senior vice provost for Research at Cornell. “Creating federated clouds has the potential to increase multi-institutional and multi-disciplinary research collaborations, enhance data-driven insights, and reduce capital expenditures.”