Leveraging the Distributed Model on Demand Platform to Work With Community Models in the Next Generation Water Resources Modeling Framework

The Distributed Model on Demand (DMOD) platform seeks to make it easier for domain scientists to use and experiment with the Next Generation Water Resources Modeling Framework (NextGen). It concentrates on several different aspects of working with NextGen, but a particularly important one is control over the necessary compute infrastructure and environments. This involves the construction and management of scalable, portable, reproducible compute environments for NextGen execution. Compute environments are complex but essential to running NextGen, and often something for which domain scientists possess more limited expertise. Thus, they can benefit from leveraging a platform like DMOD for such things.

Responsibility over compute environments brings subtle challenges. The NextGen model engine itself does not contain scientific modeling code directly. Instead it aggregates independently developed models that implement a common standard: the Basic Model Interface (BMI). This gives NextGen access to any BMI-compatible model from the scientific community, but it adds some complexity to compute infrastructure. Compute environments must ensure setup of and compatibility with any BMI modules configured for use, which may mean (depending on the module) compiling libraries, installing dependencies, and other tasks. Generally, users bear the entire responsibility for these, but DMOD can aid users by streamlining and automating such tasks.

Here we will discuss how DMOD’s design simplifies these aspects of compute environment and infrastructure management. We will first detail how DMOD builds portable, consistent job worker execution environments using container technologies, e.g. Docker. We will then discuss how DMOD leverages these technologies effectively to make it easier for users to introduce externally-sourced BMI modules - either their own or those from the community - into the DMOD job workers, without any editing of the DMOD or NextGen source code.