I work on an embedded database engine for Java. One of the things it provides are Java collections outside heap, not limited by Garbage Collector.
One of my projects is using Hazelcast. But Hazelcast is limited by GC to about 4GB of data per node. So I patched Hazelcast to use off-heap Map from my database engine. Now it handles 25GB per node without a hiccup.
It is about 3x slower, but fits 2x more data into same memory. Also it does not have GC overhead, so from about 8GB it becomes faster.
It is completely free under Apache 2 license, it is sponsored by my consulting. I would like to know your ideas what to do with this project in future. And perhaps there is similar project which could use similar optimization.
https://github.com/jankotek/mapdb-hz-offheap