There is one paragraph in this entry which needs clarification:
Java 8 brings a brand new garbage collector called Garbage First, or G1, designed for heaps greater than 4GB. G1 uses background threads to divide the heap into 1 to 32MB regions, then scan the regions that contain the most garbage objects first. Elasticsearch and Lucene does not recommend using G1GC for many reasons, one of them being a nasty bug on 32bits JVM that might lead to data corruption. From an operational point of view, switching to G1GC was miraculous, leading to no more stop the world and only a few memory management issues.
So this is unclear: should Elasticsearch clusters be using G1GC or not?
Thanks for your attention to this.