>I am running into a real bottleneck when compiling rmi applications
>with: java_dyn full version "Linux_JDK_1.1.6_v2". rmic takes forever
>to run on a single class. At first I thought it was just a jdk 1.x
>problem, but when running sun's 1.1.6 rmic finishes in about 1/30th
>the time that Linux does.
It's possible this is NFS-related; it's distressingly easy to use an
unbuffered output stream in Java. If you don't buffer writes, then
it's incredibly slow over NFS in Linux. (I think Soalris cheats and
does some write-combining, so the unbuffered writes might be faster).
Try running it so that rmic's output is written to local disk like
/tmp. Is it better? Let us know.
(This trick works with Voyager's vcc utility, which is similar to rmic
in some ways.)
