2007-05-20 09:25:37

by Chris Lee

[permalink] [raw]
Subject: A feature??

I am not a FS guru so please tell me where to get off if I sound stupid.
After lurking for some time I have come up with something that may be of
interest as a feature:

If EXT4 were to pr-allocate a configurable amount of space for a special file
and then use that file as a new EXT4 partition;
-Use that file for all new writes to disk
-Any changed files are changed by writing the change to this file, or the whole
changed file, whichever is more efficient end to end.
-A section of the file includes tracking of files that need modification in the
actual FS.
-Then when the disk is put back into normal read write mode the FS data in the
file just needs to be worked into the actual FS.

This way the FS could be put into suspended operation while a third party reads it.
I know there are other methods like snapshots, but this works at the FS level so
that the third party, which obviously needs to know ext4, can access the hard
drive directly.

Chris.


2007-05-20 11:26:04

by John Anthony Kazos Jr.

[permalink] [raw]
Subject: Re: A feature??

> I am not a FS guru so please tell me where to get off if I sound stupid.
> After lurking for some time I have come up with something that may be of
> interest as a feature:
>
> If EXT4 were to pr-allocate a configurable amount of space for a special file
> and then use that file as a new EXT4 partition;
> -Use that file for all new writes to disk
> -Any changed files are changed by writing the change to this file, or the
> whole changed file, whichever is more efficient end to end.
> -A section of the file includes tracking of files that need modification in
> the actual FS.
> -Then when the disk is put back into normal read write mode the FS data in the
> file just needs to be worked into the actual FS.
>
> This way the FS could be put into suspended operation while a third party
> reads it.
> I know there are other methods like snapshots, but this works at the FS level
> so that the third party, which obviously needs to know ext4, can access the
> hard drive directly.

Sounds like a combination of loopback and union mounting, except more
automated. Not sure there'd be much benefit to implementing those concepts
redundantly at the filesystem level.