Elliot mentioned that some JVM allows for non-contiguous memory. That is neat, but sounds to me like Ed should avoid that at all costs. I'm pretty uninformed about this, but I really don't think you want to be GC'ing over non-contiguous memory if you care at all about performance. That aside, you guys got me interested in using proprietary JVMs. Come to think of it, there have been times when I probably could have benefited a lot from that kind of thing.
-- Sam On Fri, Jan 16, 2009 at 1:05 PM, Edmund Kohlwey <[email protected]> wrote: > Good questions! > > Right now one of my biggest concerns is actually memory usage. The > application frequently garbage collects, and despite my attempts to reduce > memory consumption I still find that the rate of (and penalty incurred by) > GC'ing is excessive, and may become unacceptable. I realize that there > probably won't be much that I can do to reduce GC rate, but if a different > JVM could decrease the penalty that would be of significant value. > > > Dustin J. Mitchell wrote: > >> "We should forget about small efficiencies, say about 97% of the time: >> premature optimization is the root of all evil." (Hoare) >> >> Ed wrote, "I am ... concerned about performance of the JVM...." Well, >> JVMs are very slow at some things, and very fast at others. By >> compiling with gcj, you gain in some areas (to varying degrees over >> various JVM's) while losing in others (gcj can, in some instances, end >> up compiling down to a bunch of subroutine calls to implement complex >> operations, which is a loss over optimized JIT JVMs). >> >> So out come the old saws, "what do you mean by performance" and "how >> are you going to measure it?" That's really the only way to get to >> the bottom of this question. >> >> Dustin >> >> >> > -- /** Samuel Brin [email protected] 301-922-0357 SKYPE: sbrin77 **/
