"lucene-bytebuffer" implements Lucene Directory using Direct ByteBuffers.

RAMDirectory uses byte arrays hence subject to java garbage collection. This directory avoids byte array instead it is backed by direct bytebuffers.

Java Garbage collection has matured a lot but still difficult to get optimal performance in large JVMS. Hence I feel, for objects who's life-cycle can be known beforehand need not be subjected to java GC instead can be tracked "by-hand".

Indexing is typically lives throughout life of JVM hence can be costly in terms of GC which can be easily avoided by keeping it off-heap.

It should be used only when you are indexing large (> 2GB on 64 bit JVM) number of documents in memory. For file-based index MemoryMapped Directory should be used. Refer to Lucene documentation for details.