On Monday, 08/16/2010 at 10:47 EDT, RPN01 <nix.rob...@mayo.edu> wrote: > Flashcopy does not account for any disk buffers linux still has cached and > unwritten. It will mitigate the situation where the disk is changing while > it is being backed up. > > All in all, if you're talking about running images, full-pack backups are > basically worthless.
The thing that I don't understand is why so many computer professionals are so eager to overlook the effects of caching and consistency when taking backups. It's a variation of the classic 'time-of-test to time-of-use' problem. Heisenberg comes into play, too, after fashion. :-) If you want to know the precise position (value) of ALL the bits on the disk, their velocity, as it were, must be zero. Raise it up a few quanta and you can apply it to multiple disks. Up another level and you have multiple servers. And while an in-band backup (dd) will give you a valid filesystem, there's no guarantee of consistency of file contents if the app is still running. Back in my application developer days, the app was taken down on the weekend and the in-band Time Zero DR backup was obtained. During the week, transaction logs were collected in order to replay them in the event the db needed to be restored. Apps that need to stay up while backups are being taken need to include that capability in their design. Alan Altmark z/VM Development IBM Endicott
