On Mon, 01 Dec 2003 21:06:37 EST, Vin Shelton <acs(a)xemacs.org> said:
xemacs-21.5-2003-11-29-1924
./$d/bin/xemacs -vanilla -kill 0.65s user 0.05s system 82% cpu 0.849 total
xemacs-21.5-2003-11-29-1924-mule
./$d/bin/xemacs -vanilla -kill 2.17s user 0.16s system 93% cpu 2.498 total
So 1.52 seconds extra overhead.
Not -vanilla:
xemacs-21.5-2003-11-29-1924
./$d/bin/xemacs -kill 1.25s user 0.13s system 84% cpu 1.634 total
0.6 seconds more than non-mule -vanilla.
xemacs-21.5-2003-11-29-1924-mule
./$d/bin/xemacs -kill 3.30s user 0.16s system 91% cpu 3.792 total
But here, only 2.05 seconds more overhead - which is almost exactly the same
as the 1.52 seconds overhead for -mule -vanilla plus the 0.6 seconds more that
the non-mule version took as well.....
Looks to me that once MULE gets initialized, its performance isn't that
far out of line - but there's that big 1.52 seconds OINK at startup.
Odd indeed. I'd have expected a much bigger total hit for processing the
non-vanilla version with mule, especially when your init.el is big enough to double the
startup cost. What I'm seeing there is that the CPU costs for processing the init.el
is almost a dead heat to within the /bin/time resolution, the CPU pain is all up-front.