I am working up a proof-of-concept set of tools to do backups. It's just scaffolding right now, but it's similar to the way rsync works with the option of sending the data in chunks to tape. Each backup ends up be a full snapshot of all changed metadata and files. Metadata snapshots will get you back to a point-in-time and can be discarded at any time with little ill-effect on the system. Even unoptimized perl can scan a 400GiB set of files in less than a minute generating up to 132Mb of metadata. Because of hashing of all files and metadata it also does duplicate data folding.
If it gets out of the experiment phase I'll make a post, if anyone wants to know more I would be happy to brainstorm off-list.
Cheers,
Eric
On 4/26/06,
Francisco Reyes <[EMAIL PROTECTED]> wrote:
Alan Brown writes:
> a filesystem index at each backup (this wouldn't be difficult, but would
> be overkill in 90% of sites)
I have no idea of a percentage, but I would think that the percentage of
users that can benefit from point in time is more than 10%. Any ISP
(specially anyone using Maildir directory) needs this feature.
> For our site (~20TiB of astronomical and space probe data), the practical
> alternatives to Bacula are commercial solutions with pricetags starting
> somewhat past $10,000 and going upwards from there.
Tivoly Storage Manager is probably around or under 10K.
> As it is, Bacula is winning many converts from lesser-specified commercial
> offerings
I think Bacula has great potential. Personally I plan to see how I can help
the project on weekends (the only real spare time I have).
