#FF00AA

John Siracusa on Time Machine:

The point-in-time views in Time Machine are actually sparsely populated directory trees on an external disk or server containing mostly hard links to unchanged directories, plus full copies of the few files that have been created or modified since the last backup. […]

Time Machine leverages the same file system event notification system as Spotlight in order to keep track of which files have changed. (This notification system is open to third-party developers in Leopard. Yay!) This makes the backup process much less demanding; the entire volume does not need to be scoured for changed files, grinding the disk in the process.[…]

In other words, if you change a single byte of a 500MB file, the entire 500MB file will be copied to the backup volume during the next Time Machine backup. Frequent modifications to large files will fill your backup volume very quickly.

Neowin on Vista’s Shadow Copy:

The Shadow Copy feature in Vista doesn’t create a completely new file every time a change is made, it only creates a backup of changes using a driver that tracks changes at the block level across the entire disk volume using a copy-on-write mechanism. These changes are captured in discrete shadow copies or ‘snapshots’, which are created more or less once a day. The total size taken up by the shadow copies is capped at 15% of the disk size – this will normally give you a month or so worth of shadow copies, depending on your I/O profile.

I know that most consumers will prefer the idea of (semi) backups being stored on the same volume; as for me, I’d much rather have a complete disk clone on a separate volume so I can restore everything in case of hardware failure (Leopard installation screenshots showed a “Restore” button). Plus, I have 10GB free right now on my iMac’s 250GB drive right now, and buying a 500GB FireWire drive for Time Machine will be much simpler than upgrading the internal drive. (And that would still be true if I were a PC user.)

On the other hand, not keeping full copies of modified files is a definite Vista advantage — but I don’t think it would work too well with Time Machine’s functionality: you can instantly preview any backed-up file before restoring it, which would be much slower if you had to put together all the past snapshots in order to obtain the file’s final state (a full-volume restore based on incremental snapshots ought to be excruciatingly slow).

I’m quite puzzled by the mention of copy-on-write, though. That’s block-level copy-on-write, right? Vista isn’t actually copying the whole file whenever you make a small change in a big database, for instance, is it? (Note: copy-on-write means, instead of overwriting data you save it to a new place, and update pointers after the copy is made. So that the most inconveniently-timed power failure can’t cause data loss. It’s only slower.)