Re: Package Pool Proposal

[Joey Hess <joeyh@debian.org>]

Adam Heath wrote:
> Might it be better to use more granularity?  This could lead to alot of files
> in a dir, and we all know how debbugs handles that.  Reiserfs handles that,
> but it isn't part of a standard kernel, and we don't want to depend on
> particular features of a filesystem, when we can design it right to begin
> with.

Guy's already hashing by the first letter of the package name. This will
lead to directories that have an average of 130 package subdirectories in
them on average with the packages we have today. (We will have one directory
with 921 items in it for the letter 'l'.) Compare to today: package section
directories in source have an average of 282 files in them.

I don't see anything in Guy's design to prevent changeing it to hashing by
the first 2 letters of the package name if we need to in the future, but
debian would have to have a lot more packages in it before even a few
directories approached 1000 entries.

-- 
see shy jo