Summary

The results presented above support our assertion that fine-grained address translation can be effectively used for data structures with high fanout that are less conducive for a coarse-grained scheme. At the same time, a pure fine-grained approach is not the best performing as the primary address translation mechanism in Texas because of various overheads associated with it.

One problem with using the OO1 benchmark is that the operations do not perform any real computation (unlike in actual applications) with objects that are traversed. As such, the cost of fine-grained translation is highlighted as a larger component of the total cost than it typically would be in an actual application that performs real ``work'' on data objects as they are traversed.