On Jun 21, 2009, at 12:17 PM, Charles Hedrick wrote:
The build uses Maven 2. That's a Java-based tool, so I'm using the
same code for all OSs. It fetches some modules from repositories,
but I do that first and then build in a mode that doesn't talk to
the network. Doing a second build also makes sure that there aren't
difference in what files are in cache (though the fact that doing it
from a RAM disk has little effect suggests that I/O is not an issue).
I've used the most recent Java, directly from sun.java.com, in all
cases.
The Solaris system has Centos on it, but I did this work when
Solaris was the only OS running.
If the operating systems are sharing the same physical disk, then it
is good that you used a ramdisk for comparison as zfs will not enable
write caching unless it covers the entire disk (at least that used to
be the behaviour).
A simple truss of the build process might help determine where most of
the time is spent (user or sys) and then DTrace could be used from
here to attempt to determine where the bottleneck is.
I'll leave someone more experienced in performance analysis to make
some suggestions :)
Cheers,
--
Shawn Walker
_______________________________________________
opensolaris-discuss mailing list
[email protected]
