[Linux-cluster] DDRAID vs. GNBD+MD

[gordan at bobich.net]

>>>> On a related node, was DDRAID ever stabilised
>>>
>>> In fact, it's been removed from CVS:
>>>
>>> http://sources.redhat.com/cgi-bin/cvsweb.cgi/cluster/ddraid/Attic/?cvsroot=cluster
>>
>> Ouch. I don't suppose that means that there is a reasonable alternative to
>> it? I know DRBD works for providing RAID1 without cantralized storage, but
>> is there anything supported and available that can provide any other kind
>> of RAID?
>
> Not that I'm aware of.
>
> I think your best bet at distributed storage is a distributed cluster
> file system, a la GlusterFS, Petal/Frangipani (non-Free :( ), Lustre,
> etc.

The problem with GlusterFS is that the most it seems to provide is 
mirroring (RAID1). Lustre supports only striping (RAID0) but the 
recommendation is to run it on DRBD which provides mirroring (RAID1).

DDRAID at least provided n+1, which although not n+m at least seems to 
have been a step in the right direction. I am stunned that all these 
supposed petabyte file systems only support mirroring for redundancy. It 
really does not follow that an entity that can afford a petabyte of disks 
can also afford two petabytes of disks to achieve redundancy that doesn't 
even prevent data loss in case of failure of two speciffic disks.

Gordan