March 2014 Summaries

Reservoir simulation, a critical component in oil and gas field development and production forecasting, involves modeling fluids and gases within reservoirs using sophisticated software tools, both commercial and open-source. These simulations often require handling extensive seismic data and entail significant processing demands, but modern technology, such as tNavigator by Rock Flow Dynamics, offers scalable solutions that leverage the cloud to expedite computations. Utilizing a cloud-based platform like Rescale, users can run large-scale simulations efficiently by paying for only the computational resources they use, ensuring secure data handling and reduced computation times through scalable, parallel processing. A case study demonstrated a significant reduction in computation time from weeks to minutes by employing a powerful cluster configuration. Rock Flow Dynamics, established in 2005, partners with Rescale to provide cutting-edge reservoir simulation technology to major global oil and gas companies, highlighting their commitment to delivering high-performance, user-friendly solutions.

Rescale's platform enhances the scalability of serial codes by offering workflow tools that allow engineers to optimize internal combustion engine components like the ringpack, a crucial part of the engine, despite the lack of multi-threading or MPI support in some analysis software. Using the CASE - Ring code developed by Mid-Michigan Research, engineers can simulate and fine-tune engine performance by measuring blowby—a significant concern in engine efficiency—through parameter sweeps and optimization jobs on Rescale. Engineers can adjust parameters such as discharge and leakage coefficients or geometric properties like ring static twist and groove width to minimize blowby, using Rescale's visualization tools to analyze and interpret experimental and simulated data outcomes effectively. The process demonstrates how serial codes can be efficiently scaled out, with the platform ensuring no cost-performance loss in embarrassingly parallel applications while facilitating faster data processing through visualization tools.

The Portable Extensible Toolkit for Scientific Computation (PETSc) is a versatile suite developed by Argonne National Laboratory for solving scientific applications modeled by partial differential equations, widely recognized for its effectiveness in parallel numerical computations. Traditionally, testing new parallel algorithms developed with PETSc on a multi-core computer cluster involved extensive setup and potential administrative challenges; however, using Rescale simplifies this process by allowing scientists to specify hardware and run jobs via an internet connection. The text describes a scalability test conducted on Rescale using a modified KSP algorithm from the PETSc tutorial, which solves a linear system in parallel and logs timestamps for analysis. The test, which used varying numbers of cores, demonstrated that parallel KSP scales efficiently with more cores up to a point, as execution time decreased until 16 cores but increased with 32, attributed to randomness in matrix and vector generation affecting iteration counts. Overall, the analysis showed that the KSP algorithm performs well in parallel execution, and users are encouraged to replicate these tests with different configurations on Rescale.

Intrigued by the properties of graphene, Robert Combier, a front-end developer new to molecular dynamics (MD) simulations, embarked on a weekend project to simulate graphene using the LAMMPS analysis code. With an interest in materials science, Combier chose LAMMPS for its active community and comprehensive documentation, successfully installing it on his OS X system without needing a Linux environment. He validated his setup by running a basic tutorial simulation and later moved the experiment to the Rescale platform for enhanced performance. Utilizing a setup from an academic paper, Combier scripted his own graphene model to study the effect of altering conditions like temperature on the physical properties of carbon atoms. The project not only allowed him to visualize potential energy changes over time but also demonstrated the computational efficiency of using Rescale to reduce simulation time. Combier's experience highlighted the creative potential in designing MD experiments, particularly in exploring imperfections in graphene structures and their impact on stability.

Phil Alexander has joined Rescale as vice president of business development, bringing extensive experience from his previous role at IBM, where he managed global independent software vendor strategic alliances and distribution partners for technical computing. Residing in the Bay Area, Phil has held channel management positions at prominent companies like Cisco, Topspin, Cobalt, Netcom, and Apple. His expertise in strategic alliances makes him a valuable addition to the Rescale team. For more information about Phil and other team members, visit rescale.com/about.