Re: suggestion: do not keep files in tar archives.

blackie-1tkZHNPFh85/[email protected] (Jesper K. Pedersen) writes:

| Hi.
| 
| Yesterday I needed to restore a file from my hdup for the first time, and
| that thought me something new, namely that things can only go too slow when
| restoring.
| 
| Unfortunately it went way to slow for my - at the moment - temper.
| That made me start doing a bit of math.
| 
| I back up all of my hard disk, which includes all my digital photos etc, so
| the total backup is approx 32Gb for the base backup.
| 
| I do the backup on a IDE disk connected though a USB wrapper device, which
| is not utterly fast, but I don't really care, the priority is that I can
| easily detach it from the rest of the system. I measured that the device
| could copy in the order of magnitude 1Mb of data per sec.
| 
| Thus to simply read though the base file, I need to wait for 32.000 sec = 9
| hours. And I do need that, as I understand tar.
| 
| 9 hours is a looooong time when you accidently deleted a very important
| file from say your web server.
| 
| My suggestion is to unpack the files on the disk when copying over,
| something along those lines:
| 
| tar <options for sender> | ssh server '(cd BackUpDir; tar xpf - )'
| 
| This would store all the files on the disk, and I could thus access them
| very fast (within secs), as I would have random access to them.
| 
| Hope you like the idea and soon find time for implementing it.
| 
| Cheers
| Jesper.
I just discussed the issue with some of my coworkers, and they suggested
that afio is capable of doing random access in the files. Thus replacing
tar with afio in hdup might be even simpler than the above suggestion.

Cheers
Jesper