Re:Concept for you
Posted: Wed Jun 22, 2005 5:08 am
I'm somehow puzzled by the time limit you impose ... why exactly do you want to force the program to fullfil the "panic-file-saving" in less than 5 seconds ?
That being said, i once had the vision of an architecture where data simply won't get lost: anything you perform on your (local) data was sent to a 'journal' component through a separate memory segment (so that unintentionnally trashing of the communication array will be unlikely) and the 'journalling' component then keep the changes, perform checkpointing, write temp files, etc.
The vision has now somehow vanished into smoke so i cannot exactly remind me of how data were actually preserved by that journal component in a memory-effifient way, but the key idea was that, since that component had its own 'sub-process' (possibly co-located with the application it monitors to avoid cache flushes, but certainly unreachable from the application itself, so that in the event of a fatal crash of the application, you can restore your document at least to the last checkpoint.
That being said, i once had the vision of an architecture where data simply won't get lost: anything you perform on your (local) data was sent to a 'journal' component through a separate memory segment (so that unintentionnally trashing of the communication array will be unlikely) and the 'journalling' component then keep the changes, perform checkpointing, write temp files, etc.
The vision has now somehow vanished into smoke so i cannot exactly remind me of how data were actually preserved by that journal component in a memory-effifient way, but the key idea was that, since that component had its own 'sub-process' (possibly co-located with the application it monitors to avoid cache flushes, but certainly unreachable from the application itself, so that in the event of a fatal crash of the application, you can restore your document at least to the last checkpoint.