Thoughts on Quotas

Samuel J. Greear sjg at evilcode.net
Tue Sep 28 13:43:02 PDT 2010


On Tue, Sep 28, 2010 at 1:06 PM, Chris Turner
<c.turner at 199technologies.org> wrote:
>
> as for the # of users discussion - in these high-uid scenarios, you wouldn't
> typically share the same UID space - but have different ones -
>

Honestly, this is silly. I know that it is done, but I certainly do
not understand it. You have a few billion available uid's, why share?
Whenever I subdivide users, be it among physical machines or pfs's or
anything else, I treat uid's/gid's as a namespace. Machine #1 gets a
few hundred thousand possible uid's starting at 100000 or so, machine
#2 starting at 300000. Even if they are only going to have 50 users,
this scales just fine and you can merge at any point without conflict.

I am not the only one who uses this approach. One example: Dreamhost
uses a global uid/gid namespace, but not ranges like above, they are
allocated sequentially or so (globally) and stored in a global
repository/database to avoid conflict.
uid=101583(evilsjg) gid=377009(pg1393224)
This lets them migrate users around between their shared hosting
machines without dealing with collisions and without the complexity of
remapping.

Just because you do not do it, does not mean that it is not commonly done.

Sam





More information about the Kernel mailing list