Distributed Multi-Experiment Workflow

Name

Distributed Multi-Experiment Workflow

Context

The pattern applies to a system with the following characteristics:

• A multi-experiment workflow plan exists that lists the predetermined actions to be performed for executing each experiment plan.

• An experiment plan exists for each experiment that lists the predetermined actions to be performed while running the experiment

• Local and remote actuators may exist to allow for moving or controlling something before, during and/or after running the experiment.

• Local and remote sensors may exist to allow for measuring something before, during and/or after running the experiment.

• Local and remote instruments may exist that contain sensors and potentially actuators.

• Local and remote robots may exist that contain actuators and potentially sensors and that execute predetermined actions from the experiment plan in an automated or autonomous fashion.

Problem

Certain predetermined actions need to be performed to run a set of experiments in serial (one after another) and/or in parallel (simultaneously). One or more experiments, i.e., their tests, are remote, incurring a potentially significant communication delay between dependent experiments or to the multi-experiment workflow controller. Local experiments may involve remote components, such as by using the Distributed Experiment Control, Distributed Experiment Steering or Distributed Design of Experiments architectural patterns. Remote experiments may involve components that are completely local to the remote location, such as by using the Local Experiment Control, Local Experiment Steering or Local Design of Experiments architectural patterns.

Forces

Only pre- and post-experiment conditions are considered in performing the predetermined actions to run the set of experiments in serial and/or parallel. Safety-related conditions during the experiments may be considered. Only pre-experiment conditions are considered in performing the predetermined actions while running each experiment individually, unless the Experiment Steering strategic pattern or the Design of Experiments strategic pattern are being used for some or all experiments. If the Experiment Steering strategic pattern is being used for a particular experiment, then changing conditions during this experiment are considered in performing the predetermined actions while running it. If the Design of Experiments strategic pattern is being used for a particular experiment, then post-experiment conditions are considered in performing the predetermined actions to run it with different experiment plan parameters.

Individual experiment execution may be impacted by a significant communication delay to a set of remote experiments it depends on. Multi-experiment workflow scheduling, execution and completion may be impacted by a significant communication delay between remote experiments and the multi-experiment workflow controller.

Solution

A multi-experiment workflow controller orchestrates the execution of the experiments using a predetermined multi-experiment workflow plan (Fig. 20). The multi-experiment workflow plan’s execution is automated, performed in an open loop control and may involve human interaction. The multi-experiment workflow controller may monitor one or more experiment controllers for dependency reasons. The multi-experiment workflow plan contains a complete description of the predetermined actions to be performed for orchestrating the execution of the experiments including any dependency-related responses.

Fig. 20 Distributed Multi-Experiment Workflow architectural pattern components and control/data flow

Multiple local and/or remote experiment controllers execute their experiments using their predetermined experiment plan. Each plan’s execution is automated, performed in an open loop control and may involve human interaction. Each experiment controller may monitor the experiment for safety reasons. Each experiment plan contains a complete description of the predetermined actions to be performed for running its experiment, including any safety-related responses.

Some experiments may be executed in parallel, as they do not depend on each other, while other experiment may be executed in serial due to dependencies. The orchestration of the execution follows a DAG with the experiments as vertices and the edges as dependencies (Fig. 21). A dependency between experiments may arise when one experiment needs the result of another. Some experiments are remote in the DAG.

Fig. 21 Example of a Distributed Multi-Experiment Workflow architectural pattern directed acyclic graph

This pattern offers an open loop control with safety-related feedback on each experiment and a separate loop control with safety-related feedback for each experiment. Experiment plan execution is automated within the open loop control for each experiment, i.e., its list of actions is performed without external or human intervention that can unnecessarily hold up execution. Multi-experiment workflow plan execution is automated within the open loop control for all experiments, i.e., its list of actions is performed without external or human intervention that can unnecessarily hold up execution. A set of serial and/or parallel experiments is controlled. Some experiments are remote and there is a significant communication delay between dependent experiments or to the multi-experiment workflow controller.

Resulting Context

Experiments are executed automatically in serial and/or parallel using a predetermined plan locally and remotely, i.e., with significant communication delay between dependent experiments or to the multi-experiment workflow controller.

Related Patterns

This architectural pattern relies on the Experiment Control strategic pattern for automatically executing each predetermined experiment plan. This architectural pattern can be extended using the Experiment Steering strategic pattern (instead of the Experiment Control strategic pattern) for autonomously executing some or all predetermined experiment plans, with each plan’s parameters changing autonomously during experiments based on progress. This architectural pattern can also be extended using the Design of Experiments strategic pattern for autonomously executing some or all predetermined experiment plans, with each plan’s parameters changing autonomously between experiments based on results. The Experiment Control, Experiment Steering and Design of Experiments strategic patterns can be used together in conjunction with this strategic pattern, individually for each experiment of the multi-experiment workflow. However, the Experiment Control and Experiment Steering strategic patterns are mutually exclusive for the same experiment, as the Experiment Steering strategic pattern extends the Experiment Control strategic pattern.

In contrast to this pattern, the Local Multi-Experiment Workflow architectural pattern executes experiments that are local only, i.e., without significant communication delay between dependent experiments or to the multi-experiment workflow controller.

Examples

A science use case involving an ACL and an automated neutron diffraction experiment that further characterizes a compound, such as a grown crystal, implements a Distributed Multi-Experiment Workflow architectural pattern. The tests of both experiments are remote to each other. However, they do depend on each other as the ACL produces and initially characterizes a compound and the SNS further characterizes it. The two experiments are executed serially, requiring transportation of the compound from the first to the second.

Known Uses

This architectural pattern is used every time a set of experiments are performed, where one or more of them are considered remote. Very common examples are (1) a set of parallel experiments that investigate the same physical sample that is getting split up beforehand, or (2) a set of serial experiments that investigate the same physical sample that is getting moved from one experiment to the next with some delay due to physical distances. Each of these experiments investigates different properties, where the overall combination of the experiment results may be part of a bigger experiment that encompasses them.