Hi Mark On 20.01.22 10:18, Mark Thomas wrote: > On 20/01/2022 08:54, Olaf Kock wrote: > >> My rule of thumb is: The >> more memory there is to be claimed in GC, the longer a full GC run >> takes. > > Nope. > > The time a GC run takes is proportional to the size of objects in > memory that do not need to be GC'd. GC walks the active object tree so > it is the active objects that matter. > > Generally, the more memory you give the JVM to work with, the lower > the impact of GC for both pauses and throughput. > > A rule of thumb I have seen in the past (may be out of date for > current JVMs) is that the JVM needs roughly 5x the steady state memory > requirements for GC to work most efficiently. > Thank you for setting my ancient rule of thumb straight - noted and mentally corrected, this is really useful.
Obviously, the last time that I needed to squeeze the last bit of performance out of a machine is so old that I never needed to revert that ancient rule of thumb - now I have a cause to do so. @Lance: I take it back and claim the opposite :) Olaf --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org