From: Rob MacLachlan <ram@ri...> - 2003-10-28 13:23:21
I last tuned fasloading many hardware generations ago, but one of the
things I found was that easily 50% of the time for loading into a fresh
lisp was due to pageins for package data, etc. In orther words, due to
the linking functions of loading, and not simply getting the data in. I
would guess that if anything, hardware changes since then have increased
this effect, since disks haven't gotten faster at the same rate as CPUs.
Two things that this means:
1] Unless you can completely reconceptualize the linking aspect, this
is the ultimate speed limit.
2] You've got to be very careful in testing to get reproducible
elapsed-time results. CPU time will be more reproducible, but fairly