August 2014 Summaries

A self-contained Cloud Service package has been released to simplify setting up an InfiniBand Microsoft MPI cluster on Azure PaaS without requiring the full HPC Pack installation, addressing the challenges mentioned in previous discussions. This "MPI-in-a-box" solution allows users to deploy clusters for temporary tasks by downloading and deploying a pre-built package and adjusting a few settings in a configuration file. The package includes a startup script that installs Microsoft MPI, configures firewall ports, creates a machinefile for mpiexec calls, and installs a Cygwin OpenSSH server for remote access. Users must specify key settings in the configuration file, such as the number of instances and SSH public keys for admin and job users. Once configured, the service can be deployed through the Azure Management web page or API, enabling SSH connectivity to role instances via specified ports. The script also sets up an SMB file share for file distribution among nodes, though using azcopy for input file downloads may improve performance. The Cloud Service's source code is available on GitHub for further development and contribution.

Rescale offers a streamlined process for conducting design of experiments by providing flexible methods for variable specification and template generation, allowing users to define variable ranges via CSV file uploads or directly through a browser interface. Users can create templates with placeholders for variables, formatted as "${x}", which Rescale automatically replaces with corresponding values from a CSV file, accommodating both numeric and non-numeric data. The platform also allows users to impose specific formatting rules on their variables, ensuring consistent number formats across templates, and supports scientific notation for numerical data. This system aims to enhance the efficiency and accuracy of experiments, as overseen by Adam McKenzie, Rescale's CTO, who has a background in mechanical engineering and experience at Boeing.

Computational Fluid Dynamics (CFD) has significantly evolved as a discipline, becoming integral in industries such as aerospace and automotive for enhancing product design through simulations that complement empirical testing. Traditional second-order spatial accuracy methods, while robust and affordable, are sometimes insufficient for complex flow problems like vortex-dominated flows and aeroacoustics, which benefit from high-order methods. This text describes the application of high-order simulations using PyFR, an open-source framework, to simulate flow over a NACA 0012 airfoil using Rescale's cloud-based HPC infrastructure. By utilizing the Flux Reconstruction approach and running simulations on GPU clusters, PyFR demonstrates the capability of high-order schemes in improving accuracy without substantial computational cost increases. The simulation involved curvilinear mesh elements and advanced time integration methods, showcasing the advantages of high-order CFD simulations on modern computing architectures. Rescale's platform allows users to efficiently run such simulations, suggesting a shift towards more sophisticated hardware utilization for CFD, aligning with ongoing advancements in the field.