[Gnu-arch-users] Re: more speed improvements

[Miles Bader]

OK, I updated hackerlab too, and that improved things more, especially
the user cpu usage.  Now it seems to stand at roughly 5x faster than
without the inode-state info.

I also tried to measure the hit on the disk cache resulting from the
`what-changed', and of course that shows a much more dramatic
improvement:  in emacs, a what-changed w/ inode-state uses about 1/10th
the disk-cache vs. without inode-state, and in tla, about 1/50th!
So even emacs will be much more likely to stay cached in memory (in
which case what-changed is usually super fast).


                                                 Timings[1]
                  Counts[2]         NO inode state        W/inode state
Tree    Size  files xtgs plogs   cold hot  ucpu cache  cold hot  ucpu cache
-----   ----  ----- ---- -----   ---- ---  ---- -----  ---- ---  ---- -----
tla      4MB    451   10 1,224    33s  2s  1.6s  23MB    5s  2s  1.9s  .4MB
emacs   77MB  2,138  391    66   104s  5s  3.8s 155MB   19s  3s  2.3s  13MB

[1] cold = disk cache flushed
    hot = source tree in disk cache
    ucpu = user cpu
    cache = increase in disk cache usage after first what-changed cmd
[2] files = number of source files according to `tla inventory -s'
    xtags = number of explicitly tagged files
    plogs = number of patch-logs (NOT included in `files')


-Miles

,timings.1
Description: tla timing results [take 2]

-- 
The secret to creativity is knowing how to hide your sources.
  --Albert Einstein