better disk reliability on a desktop machine

[Alex Zbyslaw]

Chuck Swiger wrote:

> Nick Barnes wrote:
> [ ... ]
>
>> 3c: Opinions on using firewire hard disks for this at all?  Would I be
>>     better off writing DVDs?
>
>
> Hard drives provide near-online backup, but only a single full 
> iteration.  You can do incrementals to DVD or CD-RW or tape, and keep 
> many iterations handy, which is far more reliable.

A recent message on this list was from someone detailing the lengths 
they went to to prevent DVD backups from becoming unusable.  A search on 
DVD ought to find it.  Mind you, I have heard people say that DAT is 
unreliable whereas (fingers crossed) it has proved fine for me.

>
>> 3. making backups.
>>
>> 3a: I'm used to dump/restore, but it seems to me that rsync might be a
>>     better tool for this, as it would allow me to mount and browse the
>>     backup.  Opinions?
>
>
> This is good if you set up an entire system as a backup, although you 
> could dual-purpose that box and have it act as a fileserver, proxy 
> server, who knows,   as well.

I was planning something along these lines as well.  My intention is to 
have an oldish box that I can rsync to at regular intervals (probably 
from filesystem snapshots) in such a way that this would a) provide data 
backup b) provide machine backup as well.   In the meantime, it can be a 
web server or a gateway or whatever.

Originally I was going to run a couple disks with hardware RAID 1, since 
the motherboard has twin SATA RAID controllers.  But I think I'm 
changing my mind.

I've always been a bit dubious of the advantage of RAID 1.  Starting 
with two identical disks which came off the assembly line possibly 
within minutes of each other, then assuming that one fails, I believe 
that the odds of the second one also failing are greatly increased.  And 
ghods forbid, the disks you get turn out to be the next Deskstar 60 (or 
was it 75?).  Then there is the chance of controller failure.  And then 
there's the knowing if one of your RAID 1 disks has actually failed.  
Unless there is a CLI for your RAID, or FreeBSD knows enough about it, 
one disk could fail and you might not even know it, especially if you 
don't reboot regularly, or don't watch the machine POST.  On most 
desktop machines, you're stuck with one disk activity LED, which is no 
help.  Even one LED per controller isn't good enough.

So my new plan is to have two disks running RAID 0 and to rsync them 
regularly to a different kind of disk which isn't raided at all and 
which is on a different controller, as well as to the remote machine.  
If one of the raided disks fails, then I lose some amount of work, 
depending on how often an rsync is practical.  I'm prepared to live with 
that risk given that I think RAID 0 will give great benefits in some of 
the long-winded, disk-intensive, database-y stuff I do.  No doubt 
someone can tell me the error of my plan :-)  So far, it is all theory.

This is in addition to tape.

David Kelly wrote:

> There  are a few select files on the root filesystem which are unique 
> to  your system, everything else exists elsewhere such as on your  
> installation CDROM.
>
> When you go to build your new filesystem keep a list of the files you  
> tweak. Suggest placing it in /root/important_file_list. Be sure to  
> list the important file list in your important file list.
>
> tar -cvzf /home/myaccount/backups/today.tar.gz -T /root/ 
> important_file_list
>
> Size /usr sufficient for OS and application space but don't place  
> critical data there. Make /home your redundant mirror and put  
> everything critical there. 

Can't argue with the principle.  Don't forget that there are system 
specific files on /usr/local as well.  Most of it comes straight out of 
ports but there there are the config files, tweaked startup files, 
scripts in /usr/local/bin etc.  Also, if you don't have a list of the 
ports you have, then /var/db becomes important as well.

--Alex