Advanced Computing in the Age of AI | Friday, March 29, 2024

Intelligent Light Announces Kombyne HPC Efficiency Multiplier 

RUTHERFORD, N.J., Aug. 5, 2021 -- Intelligent Light announced today the creation of Kombyne an innovative new SaaS High Performance Computing (HPC) workflow tool, initially developed for customers in the defense, automotive, aerospace industries and academic research. It allows users to subscribe to a range of workflow solutions for HPC CFD jobs, from on-the-fly extract generation and rendering to simulation steering. Interactive monitoring and control is also available, all with minimal simulation disruption and no reliance on VTK. The need for large files is eliminated via extract workflows and real-time visualization.

“Over the past decade, the convergence of accessible HPC and high-fidelity simulation has created a bottleneck in analysis workflows: writing, managing and reading very large files. Intelligent Light has provided in situ solutions via VisIt/Libsim used to access information directly from the memory of the running solver code,” said Steve M. Legensky, President and CTO of Intelligent Light. Steve continued, “Based on our experience and customer input, we built Kombyne from the ground up to simplify the integration into solver codes and minimize the runtime impact on memory and performance via true ‘in transit’ operation.”

An in transit workflow uses a separate process, called an Endpoint that quickly receives data from the solver and performs visualization and analysis without interfering with the running solver. The Endpoint can directly output extracts such as cutting planes, obtain point samples for data science, render images, and it can act as a bridge to popular visualization codes such as FieldView, VisIt and ParaView, enabling interactive visualization without stopping the solver code. Kombyne is the first commercially supported product that provides full in situ and in transit post-processing, rendering and data science sampling.

Intelligent Light’s Vice President of Research and Development, Dr. Earl P.N. Duque, has been supporting a data science project, using modal and frequency analysis to characterize transonic ‘cylinder in crossflow’ physics, using NASA’s OVERFLOW2 solver, instrumented with Kombyne on the Department of Energy Cori supercomputer. According to Dr. Duque, “This study required data at high frequency, essentially every solver time step. Using the standard I/O pipelines in OVERFLOW2 would have required excessive wall clock time. Kombyne was able to efficiently write the solution extracts to disk in various data formats that could be used by data analysis tools (Matlab) and visualization tools (VisIt, FieldView, ParaView).”

Installation is extremely easy for Kombyne using prebuilt modules for OVERFLOW2, Hydra, PyFR, OpenFOAM, and more coming every day. For other CFD solvers, a small adapter library is used to link Kombyne into the solver. Kombyne provides native APIs in FORTRAN, C/C++ and Python and is scalable via MPI. In batch operation, a simple parameter datafile sets up the functions to be performed during simulation.

Kombyne comes in two flavors: Kombyne and Kombyne Lite and is supplied via an ‘enterprise open source’ site subscription, with no limit in the number of runtime instances. Kombyne Lite is available at no-cost for non-commercial use, with pricing for commercial use based on site size. Availability: Q4-2021. For more information, see kombyne.net.

About Intelligent Light 

Founded in 1984, Intelligent Light developed FieldView which became one of the world’s leading software tools for visualization and post-processing of CFD. In October 2019, Intelligent Light spun-off FieldView into an entity focused on the packaged software business (FieldView CFD, Inc.) which now allows Intelligent Light to focus on subscription-based HPC and digital twin products. The company also participates in contracted R&D efforts. Development of VisIt Prime and Kombyne was supported by the US Department of Energy via grants DE-SC0007548 and DE-SC0018633, respectively.


Source: Kombyne

EnterpriseAI