🚀 Feature (Chorish): Improve Model Runtime with Hardware and Configuration Suggestions

[DavidOry]

User Story

As a model user, I would like a full model run to be complete within 16 hours. This is the standard set forth in the Acceptance Criteria, number 30. I would also like the option of knowing what the trade-offs are for strategies that increase or decrease runtime.

Progress:

• Sufficiently defined
• Approach determined
• Tests developed
• User story satisfied
• Doc strings
• General documentation
• Passing tests

Priority

Medium

Level of Effort

Defined in Task Order 6

Resolution Ideas

At the end of development Sprint 3, the model runtime was about 70 hours (details here), with nearly all the time devoted to three activities: the resident passenger demand component, roadway assignment, and transit assignment. Each of these tasks scales nicely with additional compute resources. The current work to all parallel assignments (#175) also allows additional compute to be expended. It is therefore useful to understand what the runtime of the current system would be on different hardware configurations.

To inform decisions about precision and/or policy sensitivity, a markdown document will be created to include in the on-line documentation that provides information about configuration changes that can improve runtime. For example, the highway assignment convergence criteria can be relaxed, but doing so reduces the precision of the roadway volumes and assignment. Per @lmz, the documentation should be accessible via the tm2py repository (related: #148). Per @gregerhardt, it would be useful to have some test runs with different configurations on a relatively small project, to understand the impact empirically of relaxing precision.

Project

MTC, Task Order 6

Who should be involved?

Users: @FlaviaTsang, @Ennazus (once on-board)
Reviewers: @lmz, @gregerhardt, @DavidOry

Risk

Low. Some manual configuration will be needed until #178 is complete and testing for #175 is needed. But simple configuration changes should allow for larger runtimes.

Tests

We need to document the runtime on each machine configuration.
We need to document the performance of the test runs that inform the configuration recommendation documentation.