Re: [Bacula-devel] feature request

[Kern S. <ke...@si...>]

On Thursday 31 July 2008 16:32:49 Josh Fisher wrote:
> Kern Sibbald wrote:
> > On Wednesday 30 July 2008 18:27:42 David Boyes wrote:
> >>> I initially suggested RAIT to the submitter, but he said he was not
> >>> considering that -- if I remember right RAIT actually requires
> >>
> >> something
> >>
> >>> like
> >>> 5 tape drives. Maybe I am wrong on that though, because I looked at
> >>
> >> the
> >>
> >>> code
> >>> about 5 years ago.
> >>
> >> It's like RAID -- you need one more drive than you want to use to store
> >> data. The current LLNL RAIT library code supports 2, 3, and 5 drive
> >> stripes (1+1, 2+1, and 4+1 data+parity). It wouldn't be hard to expand
> >> that to larger stripe sizes if need be, but probably not all that
> >> interesting unless done as described below.
> >>
> >>>  In any case, once you are talking about an LTO4 drive, it is
> >>
> >> currently
> >>
> >>> impossible or nearly so to drive it full speed (except possibly by
> >>
> >> some
> >>
> >>> high
> >>> end IBM mainframes), so why complicate things when Bacula still lacks
> >>
> >> a
> >>
> >>> lot
> >>> of nice features ... :-)
> >>
> >> True, although with multiple 10G Ethernets and RTSP you can get pretty
> >> close to making a LTO4 sweat a little....8-). FICON Express
> >> multi-pathing is a lovely thing.
> >>
> >> Something like this would be more profitably be implemented as a virtual
> >> volume storage manager independent of a tool like Bacula. Hmm...
> >
> > Well, we are almost there (to your Virtual Storage Manager) ...   :-)
>
> Nothing against RAIT, but I get the impression that rather than a
> RAID-like tape subsystem, what is needed is a LVM-like subsystem for
> tape. The key word is 'subsystem', meaning that maybe it is not
> something that should be built into Bacula, and indeed would be best
> implemented as a device driver, just as lvm and md are. For example, one
> can easily create a LVM volume with striping across two or more RAID1 md
> devices and use that LVM volume with Bacula exactly the same as an
> ordinary disk drive partition. With RAIT and a LVM_for_tape driver, the
> resulting logical tape device could be used with Bacula exactly the same
> as an ordinary SCSI tape device.
>
> Just my opinion, I realize, but my opinion is that this low level device
> handling should not be built into Bacula, allowing Bacula to focus on
> what it does so well, the actual backup of networked machines. 


> Besides,  something like this is a huge undertaking, deserving of its own 
project.

Whatever way such a project is done, as a separate driver or as part of 
Bacula, I agree with the above sentence, and it will require additional 
programmers to implement it.

Kern