Thoughts on Quotas
Samuel J. Greear
sjg at evilcode.net
Sun Sep 26 09:06:13 PDT 2010
On Sun, Sep 26, 2010 at 8:56 AM, Justin C. Sherrill
<justin at shiningsilence.com> wrote:
> On Sun, September 26, 2010 8:32 am, Michael Neumann wrote:
>> My usage scenario is a Fileserver with many users. One thing that comes
>> to mind is to spend one PFS per user home directory. IIRC, HAMMER
>> supports up to 2^16 PFS. In my case it is obvious that limiting the real
>> diskspace used by each PFS would be a nice feature, not only in my case,
>> but also imagine one PFS per vkernel or jail, which would give much
>> stronger separation guaratees than what we have right now.
> The one thing that comes immediately to mind is scale. Usually, a
> DragonFly system has less than 10 PFSs. (What's the plural of PFS?)
> /home, /var, /usr, etc. Having a per-user PFS setup seems appealing
> because then you can set up snapshot and mirroring properties on a
> per-user basis.
> However, what happens where there's 200 users on a machine? It would be
> very frustrating to keep track of the settings of each user and to
> implement changes on each, especially if some users get different settings
> than others. I think the user->PFS idea is good; it just may require a
> different set of tools.
2^16 is nowhere close to enough. Quota's are very useful and to be
useful you need to support hundreds of thousands of users. I consider
lack of useful quota's to be one of our most major deficiencies.
If a person wanted to do accounting of space usage of the historical
data for snapshots, that could be done too with a userland tool by
hooking into the mirror-stream ioctl's (I don't know what indication
there is of free'd space, but I assume something could be worked out),
but I don't see that as falling under the same umbrella as quota's
More information about the Kernel