(652f) Veranova Journey to High Performance Computing on the Cloud – Azure HPC on Demand (OpenOnDemand platform)

Veranova specializes in manufacture and development of specialist and complex Active Pharmaceutical Ingredients (API). We utilize High Performance Computing (HPC) to reduce time to market for products. CFD is utilized mainly to understand and resolve issues with Mixing of heterogeneous reactions, Crystallization and Spray Drying. There are significant benefits of cloud computing including access to high CPU, GPU nodes, faster computational time when needed without upfront investment. However, the journey to transfer on-prem infrastructure is quite time intensive with heavy involvement of outside consultants to design HPC architecture. These customized solutions could also be sub-optimal design depending on the skills/experience of the consultant and the user experience.

We evaluated different cloud service providers (Azure, AWS) and secondary solution providers (Rescale, Gompute and Parallelworks). Azure was selected for the HPC production architecture in the cloud for Veranova. Azure has a significant advantage in comparison to other cloud provider in terms of low latency for bigger simulations across multiple VMs utilizing InfiniBand & full control on software installed compared to secondary providers. Azure HPC on demand is based on the OpenSource OpenOnDemand platform. Azure HPC on demand makes it easier for production deployment utilizing code to deploy architecture by simply customizing the configuration files. The deployed platform generates a user-friendly web interface. A user can quickly launch Linux desktop, code server (like Visual studio and Linux terminal), Paraview for visualization, and Jupyter notebooks for Python applications. Grafana dashboards are automatically deployed for cluster management. Computing workflow involves CAD, Meshing, simulation, run and subsequent analysis of the results. The whole workflow can be accessed using a web-interface. A hub and spoke architecture have been deployed for easy access of HPC architecture from on-prem network while ensuring network security. This HPC architecture could be easily updated to the new architecture just by redeploying with the new configuration file as newer version of Azure HPC on demand and newer hardware is released.

CFD can significantly de-risk scale-up and tech transfer to achieve “Right First Time” product. A workflow for Batch mixing case study would be used to demonstrate the significant improvement after deployment of Cloud Computing. Cloud computing infrastructure in Azure is up to 11 times faster compared to the older on-prem solution. Azure HPC allows us to run multiple simulations at the same time significantly reducing the solution time. Multiple simulations utilizing 1000s of CPU can be run in parallel to significantly reduce the waiting time for results. Simulations that would take approximately a month in-house HPC could be completed in a few days in the Azure HPC environment. This architecture can also be utilized for ML, Chemometrics, Data science and visualization to Chemical Development projects and directly deployed for end users to improve our workflow.

References

1. https://openondemand.org/
2. https://azure.github.io/az-hop/