Hi Daniel,
We are not an extremely large site userwize (actally 1/100th of your's
;-) but here's what we did anyway:
We started out by having the actual name reflect the point in the
structure where the users directory were to be found. This turned out to
be bad good solution. It did not scale. We did for instance have noone
with a name starting with x, or y, or z, and alot of people in a,
b, m, directories.
The issue is to get an even distribution accross the stucture. We found
it to be better to distribute according to the simple algoritm: "put the
next [new] user in the directory with fewest current users."
By doing this we got an even distribution (and load) accross the entire
alphabet [a-z]. It's easy to find what each users homedirectory are by
typing "echo ~user-name" now that we can't derive the home-dir location
from the name itself. Of course this works only if the user is present
locally in /etc/passwd at the node where you are running the command
which in turn calls for means to distibute /etc/passwd accross all
relevant nodes...
/peo
Daniel D. Arrasjid wrote:
>
> We're finalizing our user directory tree structure, and would like
> to know what other large sites are doing. We're thinking of
> making a directory off the top called user, and then two levels of
> subdirectories based on the first two letters of the person's
> username.
>
> For example, username 'daniel' would be in
>
> user/d/a/daniel
>
> Username 'mark29' would be in
>
> user/m/a/mark29
>
> With 60,000 users, we need a sensible way to distribute the users
> so that we don't end up with very large directories. Yet we need
> to balance that with ease of navigation for our users.
>
> Would anyone share how they manage the user space?
>
> Daniel
>
> --
> Daniel D. Arrasjid Computing and Information Technology
> Voice: (716) 645-6153 State University of New York at Buffalo
> Fax: (716) 645-5972 301 Computing Center, Buffalo, NY 14260
> E-Mail: [EMAIL PROTECTED] WWW: http://www.acsu.buffalo.edu/~daniel
> PGP public key: http://www.acsu.buffalo.edu/~daniel/key.html
--
Peo Mard
