Les Mikesell wrote at about 12:35:49 -0500 on Monday, August 31, 2009:
> Jeffrey J. Kosowsky wrote:
> >
> > > It's almost as if you guys haven't heard of filesystem-specific dump
> > > utilities. For such utils (vxdump, ufsdump, zfs send/receive, etc.)
> > the
> > > number of hardlinks isn't a problem. You can do both full and
> > > incremental dumps, even across separate machines. This isn't a problem
> > > that needs solving.
> >
> > I think you are missing some key points.
> > First, why should a program require it's own separate filesystem? This
> > seems to me like an unnatural and kludgey type of requirement.
>
> It doesn't, unless you hit some limit on the filesystem you use. The
> usual limit is how many times and how far you have to move the disk head
> since that's the slow operation.
>
> > I see lots of advantage in keeping the database portion relatively
> > small, fast, replicable, and moveable. Then you can keep and
> > distribute the files themselves wherever you want them spread across
> > one or more separate filesystems. Then the database portion is
> > optimized for what a database does best and the file-storing
> > portion can be optimized for what a filesystem stores best. And both
> > parts are easily moveable, replicable and not dependent or limited by
> > hardlinks or other filesystem-dependent functionality.
>
> But the parts aren't independent. How do you propose keeping them in
> sync or fixing them when they inevitably differ?
Perhaps analogously to the way BackupPC_Nightly now makes sure that
pool is in synch.
More generally, we would need to consider two things:
1. What are the normal ways in which the two could get out of synch
and then address each of those cases
2. If necessary, create a repair tool (similar to what I created for
the current system) if something breaks in non-standard ways (e.g.,
due to crashes, disk failures, etc.)
I guess I can't answer your question without knowing what use cases
you are worried about.
>
> > Don't get me wrong - Backuppc is great and hardlinks are a great
> > kludge to at first glance get something for nothing. I'm just saying
> > that hardlinks while "easy" bring some longer-term limitations and
> > that there comes the time when it may be worth investing in going
> > beyond them.
> >
> > Personally I would like to see Backuppc evolve to combine the pooling
> > functionality, leveraging of rsync, and relative simplicity of the
> > existing Backuppc with the expandability, portability, and flexibility
> > of the database-based systems like Bacula. I believe that the
> > combination of a database to store the file attributes and metadata
> > together with a filesystem to store the pool would be an ideal hybrid.
>
> But someone has yet to establish that this would be faster if you don't
> add the requirement of putting the sql tables on different drives - and
> weren't you just saying that applications shouldn't have requirements
> like that?
>
> > >
> > > For anyone thinking that working with giant multi-gigabyte BLOBs in a
> > > database is the right way to go, I suggest you actually attempt it
> > > yourself and see what happens. I'm backing up my HD video production
> > > rig with BackupPC, and although such a machine (Windows, 16T of
> > storage,
> > > most video files are at least 50G in size) is outside of the intent of
> > > BackupPC, it actually works. If BackupPC were to rely on an SQL
> > > database, it would greatly shrink the potential userbase.
> >
> > You are attacking a straw man. No one has ever suggested
> > "multi-gigabyte BLOBS in a database." The database would only consist
> > of the filenames, links, attrib data, and other backup-related metadata. I
> > would imagine in most cases this would be at most a couple of
> > gigabytes, assuming you have millions of files in your pool.
>
> If you don't put them in the database you can't enforce atomic
> operations - something you get in the filesystem for the price of a seek
> over to the inode to bump the link count with the entry locked.
>
> --
> Les Mikesell
> lesmikesell AT gmail DOT com
>
>
>
> ------------------------------------------------------------------------------
> Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
> trial. Simplify your report design, integration and deployment - and focus
> on
> what you do best, core application coding. Discover what's new with
> Crystal Reports now. http://p.sf.net/sfu/bobj-july
> _______________________________________________
> BackupPC-users mailing list
> BackupPC-users AT lists.sourceforge DOT net
> List: https://lists.sourceforge.net/lists/listinfo/backuppc-users
> Wiki: http://backuppc.wiki.sourceforge.net
> Project: http://backuppc.sourceforge.net/
>
------------------------------------------------------------------------------
Let Crystal Reports handle the reporting - Free Crystal Reports 2008 30-Day
trial. Simplify your report design, integration and deployment - and focus on
what you do best, core application coding. Discover what's new with
Crystal Reports now. http://p.sf.net/sfu/bobj-july
_______________________________________________
BackupPC-users mailing list
BackupPC-users AT lists.sourceforge DOT net
List: https://lists.sourceforge.net/lists/listinfo/backuppc-users
Wiki: http://backuppc.wiki.sourceforge.net
Project: http://backuppc.sourceforge.net/
|