> From: Peter Braam <Peter.Braam at Sun.COM>
> Date: June 4, 2008 8:01:20 PM ADT
> To: Peter Bojanic <Peter.Bojanic at Sun.COM>, Nikita Danilov
<Nikita.Danilov at Sun.COM> >, Johann Lombardi <johann at sun.com>
> Cc: Eric Barton <eeb at sun.com>, Nathaniel Rutman
> <Nathan.Rutman at Sun.COM>
> Subject: Re: Quotas for Lustre 2.0
>
> Customers need quota, so we need it with the DMU.
>
> There are two architectures being discussed simultaneously: one is
> quota for
> the DMU, that is new and Nikita has given a good approach. The
> second is
> adjustments to the Lustre quota architecture, which has problems,
> but it
> seems it is not a case of a re-write. So the 120 hours may not be a
> total
> waste.
>
> Nathan should have been Johann for this thread?
>
> Peter
>
>
>
>
>
> On 6/4/08 3:47 PM, "Peter Bojanic" <Peter.Bojanic at
Sun.COM> wrote:
>
>> Nikita, Peter (Braam),
>>
>> More thinking on quotas...
>>
>> Porting the existing flawed architecture from HEAD will take ~120
>> hours (be prepared to double that). It would be a shame to waste that
>> effort as a throw-away. Would the new approach that is being
>> discussed
>> on lustre-devel afford us an ldiskfs-based quotas feature on HEAD
>> before we get DMU servers? Another thought: could we have quotas
>> support for ldiskfs and disable quotas for DMU object stores if
>> necessary?
>>
>> The dependencies here are maddening.
>>
>> Thanks,
>> Bojanic
>
>
------ End of Forwarded Message