Experiments with sample applications built on top of our current infrastructure demonstrate that both morphing and fusion have significant performance benefits. Morphing provides 70% to 300% gains in performance, while fusion provides from 7.88% to 30.13% additional improvements after co-locating two agents in the same process.
The overheads caused by the acquisition of native realizations and by state transformation in morphing are relatively low, provided that morphing is not frequently invoked and that each run of the application lasts reasonablely long. We believe that the frequency of morphing will be low in most cases, as it needs to be invoked only once, unless migration is involved. However, migration itself tends to be an expensive activity; morphing will simply add some costs to this process. Similar arguments hold for object fusion.