Wow, that is quite an improvement. I wonder if we can identify any methods which can be broken up into smaller methods which would yield similiar results without needing to tweak this as much. Having perused some libraries in openJDK it is pretty clear that lots of small to tiny methods are preferred. I suspect it is for this very reason. Perhaps we can get a list of failed inlines from that wild XML generated hotspot data?
Regardless, this is very interesting. -Tom On Feb 10, 2008 5:48 PM, Marcin Mielżyński <[EMAIL PROTECTED]> wrote: > Hi, I've been trying different jvm options with rails. Turns out that > MaxInlineSize gives big advantages for rails (I've been trying it with > rails welcome-page though) > > With default options (like -J-server) the most I could squeeze from > rails welcome page was 85 requests/s after ab -n 20000 warmup. > > After a few trials I came up with -J-XX:MaxInlineSize=1000 > -J-XX:CompileThreshold=100, for first 2000-3000 requests there's been > poor 20-30 requests/s. > But after something like 3000-5000 requests ab became to show 130 > requests/s. > > With those options rails starts longer, inlines more, but ones warmed, > it can be up to 45% faster. > I know that's just a single welcome page but it already gives a measure > of lots of ruby code run. > > Marcin. > > > > --------------------------------------------------------------------- > To unsubscribe from this list, please visit: > > http://xircles.codehaus.org/manage_email > > > -- Blog: http://www.bloglines.com/blog/ThomasEEnebo Email: [EMAIL PROTECTED] , [EMAIL PROTECTED]
