Axiom: A Small Business Making Huge Strides and Bold Moves for NOAA’s HPC Cloud-based Solution for Weather Forecast Models

Silver Spring, MD, August 18, 2020 --(PR.com)-- Axiom Consultants Inc. (Axiom) announces a bold approach to get out in front of a high performance computing (HPC) strategy by developing an independently funded prototype that demonstrates Axiom’s understanding of National Oceanic and Atmospheric Administration’s (NOAA) research-to-operations (R2O) weather prediction goals and strategies. In response to a major effort of advancing toward the NOAA Unified Forecast System (UFS) and the latest initiative of establishing the Earth Prediction and Innovation Center (EPIC), Axiom, for the first time ever, ported the four-way coupled UFS Subseasonal-to-Seasonal (S2S) weather forecast application to AWS cloud HPC that mimics the prototype coupled model developed on NOAA HPC. This is accomplished in a non-typical cloud-porting approach through constructing an identical computing environment in the cloud that prevents having to train scientists, and reduces technology maintenance to easier learn techniques that scientific programmers can quickly pick up.

Axiom’s approach is not a typical porting strategy for numerical weather models where HPC can be highly beneficial in collecting and processing massive amounts of data. A typical approach to porting weather applications to the cloud is to create advanced build systems and workflow systems to replace the original, rewrite the codes to fit on different cloud architectures, and deploy it using customized technologies. While there are merits to this approach, it makes scientists' lives harder by forcing them to learn new technologies that won't necessarily improve the science, such as improving hurricane track forecasts or improving the quantitative precipitation forecast (QPF). Additional drawbacks are having to transition between configurations that only exist on the cloud, and less flexible configurations used on on-premise HPC in operations, forces scientists to port back and forth any development to operations or vice-versa. This design is stuck in the pre-virtualization-era way of thinking. Cloud HPC provides numerous capabilities and has a wide range of flexibility that can be put to better use. This is why it is imperative to have a customer-focused design philosophy.

Axiom’s design philosophy is to facilitate a simple virtualization process where software and data can simply be mirrored between an actual NOAA HPC machine and an identical cloud version. No porting by the scientist is necessary; code and scripts can be run without compilation or modification. Advanced techniques are still available, but are not necessary to use the software. This allows for scientists to be scientists and engineers to be engineers, while providing a smoother transition to the cloud vs. on-premise HPC and techniques that scientific programmers will find easy to learn.

The equivalent computing environment, along with the code and data, exists
in five layers:

The base operating system

Non-scientific software that is rarely, if ever recompiled, like Intel Parallel Studio or NetCDF

Scientific software maintained in repositories on GitHub such as the NCEP Libraries

The actual model code

Data needed to execute the model code such as input and boundary conditions

For each layer, there are options for virtualization and/or containerization, it's the developers preference. No matter what approach is taken, Layers 1 and 2 involve simply starting an operating system image followed by some installation commands. Layers 3 and 4 require checking out latest versions from the GitHub repositories. Layer 5 requires copying NOAA's famously massive datasets. As virtualization is a mandatory step in the cloud for the prototype, and containerization is an add-on, implemented iteratively to produce a variation of solutions. This prototype is the first phase of the initial capability.

Future plans entail ensuring this solution is cloud agnostic by successfully porting it to Azure and Google cloud, and eventually IBM hybrid cloud. There are plans to fully containerize the entire S2S application which includes not only containerizing the atmosphere component (FV3 GFS), but also the coupled components such as ice (CICE5), ocean (MOM6), and waves(WW3). Axiom successfully ran the containerized operational GFS with 51 nodes on AWS with an 8 minute per day log time compared to 116 nodes on NOAA operational computers; the next progressive step is to run the fully coupled S2S model containerized.

Providing NOAA with an initial cloud-based solution, design, and prototype for the vision of EPIC is the challenge Axiom has embarked on. This is an effort to simplify and improve the US’s operational weather prediction capabilities by accelerating scientific research and modeling contributions through continuous and sustained community engagement. Axiom is vested in NOAA's future to enable EPIC to accomplish this transformation with their cloud development environment; code repository; observations and tools; and community support and engagement to produce the most accurate and reliable operational modeling systems in the world. Axiom is a NOAA Weather Ready Nation Ambassador, Women-Owned Small Business (WOSB), Small Business Administration (SBA) 8(a) certified management and Information Technology (IT) consulting services provider headquartered in Rockville, MD, and is small but are making huge strides and bold moves in the advancement of global weather model prediction for NOAA.