Evaluating and improving remembered sets in theHotSpot G1 garbage collector
Abstract: In this master’s thesis project the remembered set implementation in Java HotSpot’s implementation of G1 is evaluated. It is verified by benchmarking that using Bloomfilters can reduce region scanning times when used as an intermediate data structure between card-precision bitmaps and region coarsening. It is shown that iterating the Bloomfilter is made faster by combining binary trees with the Bloom filters. It is also verified that using a more narrowinteger type, with an added bitmap to keep track of null entries, will decrease the memory footprint caused by remembered sets. Both modifications to the current implementation cause application through put regressions in SPECjbb2013.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)