[Gnu-arch-users] Re: give us a hand with arch

[Miles Bader]

"Mark A. Flacy" <address@hidden> writes:
> Miles> I've done somewhat precise calculations of the potential space
> Miles> savings, and it's quite substantial on some trees, especially those
> Miles> with lots of smallish files.  Time-savings probably would be less
> Miles> noticable, because of the recent inode-state-caching speedups.
> 
> My belief is that the time-savings would be *negative*, which is one reason
> why I'm extremely skeptical about the entire mess.

Why do you think that the time-savings would be negative?

For the no-cached-inode-state case, that seems obviously false, as
reading lots of little files is just about always a lot slower than
reading one still-pretty-little file; remember, ids are _short_, you can
probably pack about 80 of them into a _single_ disk block on a typical
ext2 filesystem.

For the cached-inode-state case, it's less clear, but as far as I can
see you still win with one-.arch-ids-file-per-directory.  If one file in
a directory changing means that you probably have to read the .arch-ids
for that directory -- but you'd have to read the .arch-ids/foo.id file
for that file in the many-.id-files case too, and if _more_ than one
file in a directory changes, the one-.arch-ids-file-per-directory case
starts to win big again.

Am I missing some obvious point???

> Which filesystem type were you using?  ext3?  

ext2/ext3

[Yes, they suck in many ways, but they're pretty much a fact of life for
the near-term future]

-Miles
-- 
I'd rather be consing.