[WBEL-users] Errors mounting raid device
Tim Moore
whitebox@nsr500.net
Tue, 06 Apr 2004 13:17:54 -0700
Michael Squires wrote:
>>As for backup I use rsync to disks on a separate machine. I will never
>
> go
>
>>back to tape.
>
>
> This doesn't protect you against file corruption that is not spotted
> before the backup disk(s) are updated (unless there are multiple copies
> of the filesystem on disk), which can be a consequence of a cracker
I don't know of any archive application or process that runs fsck prior to
making a backup. I don't worry about file corruption with a journaled file
system (ext3) nor viruses because I don't run a Microsoft operating system.
Perhaps I misunderstood?
> attack or virus infestation. It also doesn't allow for off-site storage
> or archiving of data to meet legal reporting requirements.
I didn't assume an enterprise context since there were no references
mentioned, nor are there any legal requirements for off-site storage in the
corporate world. There may be requirements to produce business records,
but no one cares how you do it as long as you can do it.
> I'll agree that tapes are expensive and tape drives even more expensive
> (and often agonizing slow), but there are applications where they're
> needed.
Nonsense. Applications don't know or care about the backup medium. I used
to backup on a Travan-5. Carts were $35USD per 20GB and very reliable,
however as my working data set grew there were several problems:
1. Speed. Backups required a quiescent filesystem (machine is off line)
and started taking hours.
2. When I did finally suffer a complete disk failure, the restore time for
a 1 week old base and 6 incrementals also took hours, not including the
hardware setup.
3. When the Travan drive mechanism broke down, I had no backup or restore
until I found a replacement of the same type.
Switching to disk-based backup and rsync provided the following:
1. Backup time takes 95% less time than the Travan. Backup data rates are
about 9.4MB/s on a 100bT network. If I had GigaBit, they would be at the
backup disk's write speed.
Backups are still run during the most quiet time but do not require a
quiescent filesystem, tape change or any other operator intervention. No
more scheduled downtime.
2. I have an exact replicate of the filesystem hierarchy on the backup
disks. Each machines backup is exported back to itself readonly with the
same file permissions. No worries about proprietary tape or backup formats.
3. Incremental snapshots are done hourly with hard links used for files
that haven't changed. This means the date/time stamped incremental trees
are also 100% complete replicates of the machine's entire filesystem, yet
use only the physical storage space of the actual changed files.
Since rsync uses a rolling checksum and block replacement algorithm, only
the changed blocks for modified files are sent between the source and
backup. Changing just a few blocks of a 500MB file results in an
incremental data transfer of just those few blocks rather than another
500MB file.
4. I am not tied to any particular hardware requirements for disks or disk
controllers. In fact I recycle my older, slower 5400 RPM EIDE drives and
old Promise Ultra/33 controllers.
5. The rsync application I use is open source and so I have the source code.
rgds,
tim.