
Signature
What I write is what I mean. I request that anyone who decides to respond
please refrain from "disagreeing" with something I didn't write in the first
place.
>> I've gotten a bit weary of the complexities and long execution times of
>> Retrospect backups -- even incremental backups seem to involve a very
[quoted text clipped - 4 lines]
> changed since the last backup. Which means looking at the directory
> entry for every file on every volume that's scheduled for backup.
Well, if Dantz was a better vendor, they would probably already have
fixed this issue for Macs running Tiger. Now that Spotlight has arrived,
generating a list of all files modified since a particular point in
time should be a simple query to the spotlight index.
Full scans would still be required for any volume that contains any
folders in the "Privacy" mode.
>> -- and am thinking of converting to weekly disk image
>> (.dmg) backups instead.
[quoted text clipped - 8 lines]
> backups is that they let you go back to any of several previous points
> in the life of a document.
And, as noted already, if you choose to do full clones, and erase your
only copy before making a new backup, you are seriously at risk of
finding out about "soft" data corruption during the copy. It is quite
possible for disk problems to "hide" in unread portions of the drive
until it comes time to read that portion.
27GB cloned to a 40GB external definitely fits that scenario.
But this problem actually applies to all backup solutions - it's just
that the "delete, then clone" cycles more frequently than the "base
dump, then inc, inc, inc, inc, inc, ..., reset new base dump". Two
alternating backups (either clone or traditional base+inc) fixes
this problem, but at a significant additional cost of backup
space/media.
I still use Retrospect, because 1) it allows me to go back to previous
versions of files, and 2) because it allows me to back up all my systems
(Mac, PC, and Linux) on a single server.
For multiple networked systems, I think an automated network backup
like Retrospect is the only way to go. Even for single systems, the
choice between cloning and Retro isn't clear, since you can get
Retrospect Express bundled with external hard drives and internal or
external DVD-RW drive for practically free from www.macsales.com.
Gregory Weston - 23 Oct 2006 18:30 GMT
> >> I've gotten a bit weary of the complexities and long execution times of
> >> Retrospect backups -- even incremental backups seem to involve a very
[quoted text clipped - 6 lines]
>
> Well, if Dantz was a better vendor,
Note that it's only nominally Dantz any more. It's now EMC....
> they would probably already have
> fixed this issue for Macs running Tiger. Now that Spotlight has arrived,
> generating a list of all files modified since a particular point in
> time should be a simple query to the spotlight index.
It's not.
A. Spotlight's not reliably available for their customer base, and more
importantly
B. Spotlight's not reliably functional among the subset of their
customer base for whom it is available.
So there's not really much incentive for them to do so.
> But this problem actually applies to all backup solutions - it's just
> that the "delete, then clone" cycles more frequently than the "base
> dump, then inc, inc, inc, inc, inc, ..., reset new base dump". Two
> alternating backups (either clone or traditional base+inc) fixes
> this problem, but at a significant additional cost of backup
> space/media.
True. It always comes down to deciding how much your data is worth to
you. With 500GB FW drives now readily available for under $300, and
250GB drives available for under $100 to those who invest a little time
I think the cost is rapidly losing significance.
> I still use Retrospect, because 1) it allows me to go back to previous
> versions of files, and 2) because it allows me to back up all my systems
> (Mac, PC, and Linux) on a single server.
Ditto, on both counts.
G

Signature
What I write is what I mean. I request that anyone who decides to respond
please refrain from "disagreeing" with something I didn't write in the first
place.
Jolly Roger - 24 Oct 2006 00:41 GMT
>>> I've gotten a bit weary of the complexities and long execution times of
>>> Retrospect backups -- even incremental backups seem to involve a very
[quoted text clipped - 10 lines]
> generating a list of all files modified since a particular point in
> time should be a simple query to the spotlight index.
You are assuming that Retrospect only looking at the modification date
and nothing else. Bad assumption, there. If Retrospect compared only
modification dates, we'd be in a *lot* of trouble! And I doubt very
seriously shoe-horning Spotlight in would do anything but make the
comparison take much longer, because, with all the overhead, it's
simply not the right tool for the job. Your assumption shows how little
you really know about professional-quality incremental backup software.
This isn't an "issue" that needs to be fixed on Tiger - it's just the
way it works.

Signature
JR
James Glidewell - 24 Oct 2006 17:03 GMT
>>>> I've gotten a bit weary of the complexities and long execution times
>>>> of Retrospect backups -- even incremental backups seem to involve a
[quoted text clipped - 20 lines]
> isn't an "issue" that needs to be fixed on Tiger - it's just the way it
> works.
I'm glad there are truly smart folks around who can educate me then.
Tell me how the contents of a file can be changed on any Unix system
without the file modification field being set. And what sorts of file
metadata updates are _not_ captured by mdimport...
I'd also like to hear your speculation about how Leopard's Time Machine
works, given that all reports I have heard indicate that it does NOT do
a daily full volume scan.
While all the data may or may not be in place to do incremental dumps
without a full file system scan in Tiger, they are apparently there in
Leopard.
Your assumption that the metadata datastore indexes cannot _possibly_
contain the info needed to do a proper incremental dump is pretty lame
given that fact that Apple will be using those same indexes for Time
Machine.
Gregory Weston - 25 Oct 2006 13:04 GMT
> Tell me how the contents of a file can be changed on any Unix system
> without the file modification field being set. And what sorts of file
> metadata updates are _not_ captured by mdimport...
The modification date field itself can be fiddled with. So you could end
up with an unchanged file whose date has changed or a changed file with
a static mod date. I honestly have no idea how Retrospect will handle
either of those situations because I'm not sure offhand what criteria
they use to determine whether a file is in need of backup.
The sorts of metadata that aren't captured by mdimport are anything that
the single importer selected to process a file doesn't update. This can
happen by omission, intentional or otherwise, or the import process
could fail as happens with distressing regularity on my G5. Spotlight,
today, does not provide a reliable enough selector for a mission
critical task like data backup.
> I'd also like to hear your speculation about how Leopard's Time Machine
> works, given that all reports I have heard indicate that it does NOT do
> a daily full volume scan.
Bluntly, that's not especially relevant. It's a major revision of OS X
away and there's no telling how much has changed. There's a lot of room
for improvement in Spotlight over what 10.4.x has delivered. I hope, for
example, that they've decided to update Spotlight such that multiple MD
importers are given a shot at any given file.
> While all the data may or may not be in place to do incremental dumps
> without a full file system scan in Tiger, they are apparently there in
> Leopard.
Right. But we're talking about Tiger and earlier. So if it's not
reliable today, it's not a good expenditure of resources for product
that's shipping today.
> Your assumption that the metadata datastore indexes cannot _possibly_
> contain the info needed to do a proper incremental dump is pretty lame
> given that fact that Apple will be using those same indexes for Time
> Machine.
So the summary response is:
1. The behavior in Leopard is not relevant to a currently-shipping
product. When more than 0% of EMC's customers are using Leopard, that's
likely to change.
2. The problem is not that "the metadata datastore indexes cannot
_possibly_ contain the info" it's that currently they do not _reliably_
contain that information.

Signature
What I write is what I mean. I request that anyone who decides to respond
please refrain from "disagreeing" with something I didn't write in the first
place.
Paul Sture - 25 Oct 2006 16:18 GMT
> I'm glad there are truly smart folks around who can educate me then.
>
> Tell me how the contents of a file can be changed on any Unix system
> without the file modification field being set. And what sorts of file
> metadata updates are _not_ captured by mdimport...
Try the experiment in my next paragraph, and it's at least one reason
why Mac backups have to churn through everything to decide what to copy
/ back up.
I believe things changed with Tiger, but under Panther, move a file from
one folder to another using Finder. Both folders reflect the
modification date, but the file itself does not.
File contents have not changed, but it's now in a different place, and a
backup program needs to work that out.
The only way I know to achieve both fast incremental backups and data
integrity on a restore operation is to record:
a) the date and time of the backup of each and every file, including
where it resides.
b) a backup of _all_ directories, including date and time, of course.
c) the backup date and time recorded needs to be that of the backup
start, to cover files which may be modified during the execution of the
backup.
It's a more complex subject than it would seem to to be to the casual
observer.
Feedback welcome, 'cos I'd like to know how it's possible to speed up
incremental backups in an OS X environment.

Signature
Paul Sture