2.2.3 - During this cycle, QE has used these settings: -XX:+UseG1GC -XX:InitiatingHeapOccupancyPercent=70 -XX:MaxGCPauseMillis=3000. Once again, not saying these are the only "blessed settings", but maybe we should recommend customers use G1? Or at least used in the example configuration in section 2.4 which does not include a G1 example.

2.3 - The "Minimum Requirements" section talks about standalone.conf and domain.conf which only apply to the JDG Server, so I think you mention that library mode is set through the command line.

Under "Recommended Memory Requirements" the doc should describe the following rule: The amount of data in the heap should not exceed 50% when only basic operations are used (i.e. put, get, remove, etc.), and 33% of the heap when analytics (i.e. Queries, Streams, Distributed Executors, Map/Reduce, etc.) are used against the cache. The extra free memory allows the system to deal with memory allocations from these analytic operations.

Here are rhn-engineering-bban's comments:

• 4.3.1/4.3.2: I'd add a ymmv; and changes should be measured wrt perf

• 4.4: makes sense when sending data that's bigger than the MTU (e.g. 9K
  for jumbo frames), or when smaller messages are aggregated in the
  JGroups bundler (transport) to be > 9K