Scalable Armies of Model Clones through Data Sharing
Abstract
Cloning a model is usually done by duplicating all its runtime objects into a new model. This approach leads to memory consumption problems for operations that create and manipulate large quantities of clones (e.g., design space exploration). We propose an original approach that exploits the fact that operations rarely modify a whole model. Given a set of immutable properties, our cloning approach determines the objects and fields that can be shared between the runtime representations of a model and its clones. Our generic cloning algorithm is parameterized with three strategies that establish a trade-off between memory savings and the ease of clone manipulation. We implemented the strategies within the Eclipse Modeling Framework (EMF) and evaluated memory footprints and computation overheads with 100 randomly generated metamodels and models. Results show a positive correlation between the proportion of shareable properties and memory savings, while the worst median overhead is 9,5% when manipulating the clones.
Domains
Software Engineering [cs.SE]
Fichier principal
modelcloning-final2.pdf (258.85 Ko)
Télécharger le fichier
slides-final.pdf (449.57 Ko)
Télécharger le fichier
Origin | Files produced by the author(s) |
---|
Origin | Files produced by the author(s) |
---|
Loading...