Parallel
A SimulationOracleIfc that executes the points of an EvaluationRequest concurrently, each on its own freshly built Model, honoring the request's common-random-numbers (CRN) vs. independent stream choice through per-point run parameters.
A single-point request has nothing to parallelize, so it is short-circuited to one reused model (no build, no coroutine) via an embedded SimulationProvider. Point-by-point solvers therefore pay no extra cost in parallel mode; only multi-point batches (e.g. a Cross-Entropy population or an R-SPLINE simplex/neighborhood) fan out across models.
The supplied modelBuilder MUST return an independent, freshly built Model on each call to build(...) (a bundle-backed builder does). A builder that returns a shared or cached instance is a data race under concurrent execution and must not be used here.
Each point runs a full experiment on its own model, mirroring the proven ParallelDesignedExperiment pattern; per-point conversion to a ResponseMap happens sequentially on the calling thread after the parallel phase completes. Built models are kept silent (no summary printing, no CSV) and never touch their (now lazy) outputDirectory, so a long run neither interleaves stdout nor accumulates output-file handles.
Parameters
builds a fresh model per point; must yield independent instances
opaque configuration forwarded to modelBuilder.build(...)
baseline run parameters shared by all points; if null, taken from the template model
an already-built model to adopt as the template and single-point reuse model; if null, one is built
optional cache used only by the single-point (sequential) path
when true (default), size-1 requests run on the reused model instead of the parallel path
the bounded dispatcher governing worker concurrency; defaults to a dedicated view sized to the processors
Constructors
Properties
Functions
Executes multiple simulations based on the provided evaluation request and maps each request to a corresponding ResponseMap. Each request is processed individually, and the results of the simulations are stored as a key-value pair in the returned map. If the input list is empty, an exception is thrown. This default implementation runs all the requests sequentially based on the order within the evaluation request.