[Metalab] Distributed Redundant Storage

[Markus Kienast]

The implementation of btrfs has been analyzed by some RedHat dev some
time ago. And he concluded, that btrfs says it is following certain
concepts of FS design, but actually does not strictly follow these
concepts and diverts from them on impulse. These diversions have not
been scientifically tested and WILL lead to unexpected behavior and
unpredictability.

One of which is the inability to correctly predict free space in the
FS, if I remember right.

I am referring to that paper, when I say "unprofessional". This
analysis seems to have been ignored in the press mostly, which does
not surprise me, for the  knowledge level needed to actually
understand, the point the guy is trying to make.

In short he says, by doing what btrfs devs are doing, they will never
get out of the experimental state, as some unpredictable bullshit will
pop up until worlds end. And if you look, how long it has been around
and how many distributions use it as their default FS today (although
its features are clearly everything we ever wanted), this prediction
seems about right.

Cheers,
Markus

On Mon, Nov 12, 2012 at 11:10 PM, gaelic <gaelic at luchmhor.net> wrote:
> If you read the ceph mailinglist you also should know that you can use
> virtually any FS you wan't and are not bound to btrfs, btrfs is only
> suggested.
> And I don't think you mean btrfs is unprofessional ... but at the
> moment its still signed as 'experimental' by the devs. Still, my
> experience is that it's very reliable as I never had failures within
> the last 2 years on several machines.
>
> Regards,
>
> g
>
> On Mon, Nov 12, 2012 at 10:42 PM, Markus Kienast <elias1884 at gmail.com> wrote:
>> Ceph, as far as the specs are concerned, is exactly what I need.
>> But I do not really believe in its reliability, as I do not believe in
>> the professionality of its underlaying FS btrfs.
>> I am on there mailinglist and what I read is not so promising.
>>
>> Concerning the "productivity" of my system. Yes, it is production. I
>> rent a full rack at a datacenter and I have to care for storage
>> myself. I however would prefer a scalable system based on cheap
>> commodity hardware over installing an expensive enterprise SAN
>> solution.
>>
>> So long,
>> Markus
>>
>>
>> On Mon, Nov 12, 2012 at 10:12 PM, Zacharias <mail at sacharja.eu> wrote:
>>> Am 2012-11-12 21:39, schrieb elias humbolt:
>>>
>>>>
>>>> Can anybody recommend a redundant replicated storage solution stable
>>>> enough to act as basis of production system?
>>>>
>>>> It will be used to store and serve video data to the web and to
>>>> transcoding nodes.
>>>>
>>>> Any recommendations?
>>>>
>>> Yes, first of all:
>>>
>>> If it is a *production* system, this is, something you earn money with: Rent
>>> a server in a data center. It is their job to care about redundancies.
>>>
>>> If it is *not* a "production system", this is, you are basically doing this
>>> for fun, you might wanna check out ceph http://en.wikipedia.org/wiki/Ceph
>>> although I admit I never used it myself.
>>>
>>> greetings, Zacharias
>>>
>>> _______________________________________________
>>> Metalab mailing list
>>> Metalab at lists.metalab.at
>>> https://lists.metalab.at/mailman/listinfo/metalab
>>
>> _______________________________________________
>> Metalab mailing list
>> Metalab at lists.metalab.at
>> https://lists.metalab.at/mailman/listinfo/metalab
>
> _______________________________________________
> Metalab mailing list
> Metalab at lists.metalab.at
> https://lists.metalab.at/mailman/listinfo/metalab