Creation of large numbers of co-existing long-lived isomorphic objects
increases the memory footprint of applications significantly. In this paper
we propose a dynamic-analysis based approach that detects allocation sites
that create large numbers of long-lived isomorphic objects, estimates
quantitatively the memory savings to be obtained by sharing isomorphic
objects created at these sites, and also checks whether certain necessary
conditions for safely employing object sharing hold. We have implemented
our approach as a tool, and have conducted experiments on several real-life
Java benchmarks. The results from our experiments indicate that in real
benchmarks a significant amount of heap memory, ranging up to 37% in some
benchmarks, can be saved by employing object sharing. We have also
validated the precision of estimates from our tool by comparing these with
actual savings obtained upon introducing object-sharing at selected sites
in the real benchmarks.