From: <fle...@wo...> - 2006-08-05 18:12:37
|
Hi I didn't mean to be rude either,=20 it's just that having a backupsystem that really works like it "should be"= =20 just doesn't exist currently (to my knowledge). =46or you're example, (a.conf, b.conf) you need some filesystem support to = be=20 able to replicate your changes to another system/backup. Currently there's no method/sw that I know of that is supporting=20 windows/unix/linux and is in common practise=20 (otherwise it's difficult to use in an enterprise environment due to the sh= ere=20 amount of servers and different OS's). What you maybe are searching maybe FAM (File Alteration Monitor)?!?=20 http://oss.sgi.com/projects/fam/faq.html And use of a FAM mirror http://www.linuxfocus.org/common/src/article199/fam_mirror Unless you set up something like that, you're not getting any solution to y= our=20 a.conf/b.conf problem, the "old" methods" of full/incremental/differential= =20 backups just doesn't provide that functionality. Also isn't a tape backup=20 system flexible enough, or fast enough to support that. > I will not trust my system+data to such a backup-system. Are *you* > comfortable with that? What is the down side of being able to restore > reliably? Rather too much than too little data protection is what keeps an enterprise= =20 datacenter running... > You are right, I don't know "EMC Networker, Veritas Netbackup, HP Data > Protector", but I'm willing to bet they can restore a directory > containing *exactly* b.conf. (Am I right?) They can, if you really have an full backup just before the crash/logical=20 accident happens. For incrementals, it's just to restore full+incrementals= =20 and get all those a.conf's as well ;-( =2D-Robert On Fri 4 August 2006 09:50, Peter Valdemar M=F8rch wrote: > flexbackup-at-worreby.ch |Lists| wrote: > > 1. Have you the slightest idea how it's done in the reality? > > > > That's standard procedure to restore full backup plus all > > incrementals you have, if you invented some other schema, > > please let me know > > (at work we're currently backing up between 15-20 TB / night) > > and > > > 2. What's worse? Dataloss or too much data!!! > > Let us first agree that data loss is totally unacceptable. > I'm not arguing that data loss is a good thing. (Am i? Where?) > > But for me, "too much data" is *ALSO* unacceptable, because it does not > represent reality. Let me illustrate with a scenario: > > * I start with a directory e.g. under Apache's configuration that > has a single file a.conf > * I make a full backup. > * rm a.conf > * Add a file b.conf > * (Notice that at no point in time was there ever more > than one file in the directory) > * I make an incremental backup. > * Hard disk crash > * Restore full backup > * Restore incremental backup > > Result with current flexbackup: > > A directory containing both a.conf AND b.conf. > > Wouldn't you rather end up with a directory containing *exactly* b.conf > - the exact contents of the directory when the incremental backup was mad= e? > > Ok, so in a directory where there is supposed to be 1 single file you > may be able to remember yourself to delete a.conf (because it is *not* > supposed to be there). Lets just hope I slept well and didn't remove > b.conf instead by accident because I was tired... ;) > > But with "between 15-20 TB" of data as you put it? Who can remember > which of the gazillion files to delete? > > No, I don't want that to be a guessing game. > > In some (most?) cases, "too much data" is just annoying - not really > catastrophic. But in some cases (e.g. conf.d, cron.d, *.d directories, > or a file containing sensitive data that was deleted) this *is* > catastrophic. I could end up with a system that won't boot or misbehaves > or is dangerous after a full+incremental restore. > > I will not trust my system+data to such a backup-system. Are *you* > comfortable with that? What is the down side of being able to restore > reliably? > > You are right, I don't know "EMC Networker, Veritas Netbackup, HP Data > Protector", but I'm willing to bet they can restore a directory > containing *exactly* b.conf. (Am I right?) > > Peter =2D-=20 =2D-Robert Robert Worreby Birkenweg 82 CH-3123 Belp http://counter.li.org |