Config MDEO plugin

Contributes the search, solver and runtime sections: which moves the search may make, which algorithm makes them, and what resources the run may use. It also owns the executable section, so this is the plugin that actually starts an optimisation.

At a glance

Plugin id	config-mdeo-service
Display name	Config MDEO
Description	Language support for config MDEO sections (search and solver)
Default URL	/plugin/config-mdeo
Source	app/packages/service-config-mdeo, app/packages/language-config-mdeo
Depends on	The Config and Config Optimization plugins; executions go to optimizer-execution

Languages contributed

Language id	Name	Extension	Textual editor	Graphical editor	Generated
config-mdeo	Config MDEO	—	❌	❌	✅

The config-mdeo language

A generated language with no extension and no editor, used the same way as config-optimization: the config service sends it the fragment of a .config file containing this plugin's sections, and it parses them with a standalone grammar built from the same rules.

Contribution plugins contributed

Target language	Sections	Executable
config	search, solver, runtime	solver

Contribution plugin id	config-mdeo
Short name	mdeo
Grammar dependencies	config-optimization
Section dependencies	problem from optimization

The section dependency matters: before this plugin can resolve create Annotation or mutate Canvas.layers it has to know which metamodel the problem uses. The config service therefore computes the problem section first and passes its result along.

The search section

The mutation operators available to the search.

search {
    mutations {
        using "./match.mt"

        create Annotation
        delete Annotation
        mutate Circle

        add Layer.shapes
        remove Layer.shapes
        mutate Canvas.layers
    }
}

Entry	Meaning
using "path.mt"	Use a hand-written model transformation as a mutation operator
create Class	Generate an operator that adds an instance of the class
delete Class	Generate an operator that removes an instance
mutate Class	Generate an operator that changes an instance's properties
add Class.edge	Generate an operator that adds a link
remove Class.edge	Generate an operator that removes a link
mutate Class.edge	Generate an operator that rewires a link

Entries may be mixed and repeated in any order. Generated operators are written into the result tree as .mt_gen files, so you can read exactly what the search was allowed to do.

The solver section

The search algorithm and when to stop. This is the executable section: a config file containing it gets a run action.

solver {
    algorithm = NSGAII

    parameters {
        population = 40
        variation = mutation

        mutation {
            step = interval(1, 5)
            strategy = random
        }
    }

    termination {
        evolutions = 500
    }

    batches = 3
}

Key	Values	Meaning
algorithm	NSGAII, IBEA, SPEA2, SMSMOEA, VEGA, PESA2, PAES	The multi-objective algorithm
batches	integer	Run the same configuration this many times independently

parameters

Key	Values	Meaning
population	integer	Population size
variation	mutation, genetic, probabilistic	How offspring are produced
bisections	integer	PESA2 / PAES only
mutation { … }	block	Mutation tuning, see below
archive { size = n }	block	PESA2 / PAES only

mutation

Key	Values	Meaning
step	n, fixed, fixed(n), interval(lo, hi)	How many operators to apply per mutation
strategy	random, repetitive	How operators are chosen
selection	random	How matches are selected
application	random	How an operator is applied
credit	random	Operator credit assignment
repair	default	Repair strategy after a mutation

step = interval(1, 5) draws uniformly from [1, 5) on every call; fixed with no argument means one.

termination

Key	Meaning
evolutions	Stop after this many generations
time	Stop after this many seconds
delta	Stop when the improvement stays below this threshold
iterations	Number of non-improving iterations tolerated before stopping

Several conditions combine with OR — the first one to trigger ends the run.

The runtime section

Execution limits.

runtime {
    timeout {
        script = 1000
        transformation = 1000
    }

    backend = MDEO

    resources {
        threads = 10
        nodes = 4
        threadsPerNode = 3
    }
}

Key	Meaning
timeout.script	Milliseconds a single objective or constraint evaluation may take
timeout.transformation	Milliseconds a single transformation application may take
backend	MDEO (default) or Tinker — the graph representation used during search
resources.threads	Upper bound on total worker threads
resources.nodes	Upper bound on execution nodes taking part
resources.threadsPerNode	Upper bound on threads per node

All fields are optional and act as upper bounds against what the deployment actually offers.

Full example

// Which model is optimised, and against which metamodel.
// Contributed by the Config Optimization plugin.
problem {
    metamodel = "./shapes.mm"
    model = "./shapes.m"
}

// What "better" means. Contributed by the Config Optimization plugin.
goal {
    import { shapeCount, invisibleShapes, emptyLayers } from "./metrics.fn"

    minimize invisibleShapes
    maximize shapeCount
    constraint emptyLayers

    refine Canvas.layers[1..4]
}

// How new candidate models are derived. Contributed by the Config MDEO plugin.
search {
    mutations {
        using "./match.mt"
        using "./control-flow.mt"

        create Annotation
        delete Annotation
        mutate Circle

        add Layer.shapes
        remove Layer.shapes
        mutate Canvas.layers
    }
}

// Which search algorithm runs, and for how long. Contributed by the Config MDEO plugin.
solver {
    algorithm = NSGAII

    parameters {
        population = 40
        variation = mutation

        mutation {
            step = interval(1, 5)
            strategy = random
            selection = random
            application = random
            credit = random
            repair = default
        }
    }

    termination {
        evolutions = 500
        time = 600
        delta = 5
        iterations = 3
    }

    batches = 3
}

// Execution limits for the run. Contributed by the Config MDEO plugin.
runtime {
    timeout {
        script = 1000
        transformation = 1000
    }

    backend = MDEO

    resources {
        threads = 10
        nodes = 4
        threadsPerNode = 3
    }
}

Server-side capabilities

Key	Kind	Contents
ast	file data	The serialised AST of the standalone language
config	request	Computes the section data for search, solver and runtime
config-execution	request	Starts an optimisation run
config-execution-get-summary	request	The markdown summary of a run
config-execution-get-file-tree	request	The result file tree
config-execution-get-file / -get-files	request	Result file contents
config-execution-cancel	request	Cancel a running optimisation
config-execution-delete	request	Delete a finished run and its results

Execution requests are forwarded to the optimizer-execution service, which is the component that scales across nodes. See Reading the results for what comes back.