Use Cases

Identify and categorize use cases and scenarios motivating SCALE-MS design and packaging.

Application user

High level use cases for applications built on the Python client library.

Application code

Use cases encountered when implementing user interfaces or high-level applications with the Python data flow client library.

Adaptive MSM

level: algorithm
Synchronous fixed-width ensemble scenario:
1: Set up an array of N simulations, starting from a single input.
2: Evaluate “while loop” with ensemble condition.
2a: Publish MSM data if converged.
2b: Execute an iteration before reevaluating loop condition.
.1: For each ensemble member, prepare simulation input for next iteration.
.2: Perform N uncoupled MD simulation segments.
.3: Collect conformation samples across ensemble.
.4: Apply new samples to update MSM state.
.1: Update transition matrix.
.2: Calculate convergence metric.
.3: Generate N conformations for the next iteration.

Collect conformation samples across ensemble

level: scripting construct / algorithm building block

Use case description: If N trajectory segments contain M conformation samples, each, create a reference to the collection of X=N*M samples.

Scenarios: Data transformation and placement scenarios depend on how the reference is consumed.

Variant: For some data flow topologies, this equates to an AllGather step before downstream operations.

Variant: In the absence of any other facilities, this is typically accomplished for GROMACS data by performing trajcat on the ensemble trajectory files.

Scripting user

Python code written directly against the data flow client library.

Client API library

Use cases encountered in the client library implementation, interacting with the task execution middleware in support of of client use cases.

Execution middleware

Resource management, data placement, task discovery, and task scheduling use cases. Interactions include various architectural aspects, compute/data/communications abstractions, quality of service guarantees, and performance optimizations.

Generic sequential task execution

Units of work (nodes) describe the inputs and expected outputs, and identify a module able to perform the operation.

Operation implementation

Use cases encountered when implementing operation code to be executed by the runtime.