Hi David, Would be great if we could achieve that, because that means most companies would only need one server (ok + a failover instance) . Let's slowly work towards these numbers. We could even user more heap, if that would turn out to be the problem. The concurrent mark and sweep collector works very well with only minor "stop the world" phases at least up to 6,5 Gbyte.
Regards, Markus "The best way to predict the future is to invent it" -- Alan Kay On Mon, Nov 23, 2009 at 5:39 PM, David Pollak <[email protected] > wrote: > My back of the napkin sizing when I first started developing ESME was that > a > JVM with 4GB of heap could handle ~ 25,000 active users on a quad core > opteron/xeon. Active is on average 1 post per 30 minutes with an average > fan-out (followers) of 100 and peak fan-out of 2,500 (10% of the user > population). > > If this sizing seems reasonable, let's build further benchmarks around it. > > On Mon, Nov 23, 2009 at 2:29 AM, Richard Hirsch <[email protected] > >wrote: > > > Thanks. > > > > Updated the wiki page. > > > > D. > > > > On Mon, Nov 23, 2009 at 11:21 AM, Markus Kohler <[email protected] > > > > wrote: > > > Haven't checked all the details, but heap size was 1Gbyte. > > > I need to define the recommended CM settings > > > Markus > > > > > > On Nov 23, 2009 8:55 AM, "Richard Hirsch" <[email protected]> > wrote: > > > > > > Hi, > > > > > > I just finished posting Markus' last analysis from his continuing > > > performance tests: > > > > > > http://cwiki.apache.org/confluence/display/ESME/Performance+test+-+2009-11-22 > > > > > > Very interesting results. > > > > > > @Markus: maybe you can check the wiki page to make sure my editing > > > hasn't added any errors. > > > > > > Thanks. > > > > > > D. > > > > > > > > > -- > Lift, the simply functional web framework http://liftweb.net > Beginning Scala http://www.apress.com/book/view/1430219890 > Follow me: http://twitter.com/dpp > Surf the harmonics >
