[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

RE: searching for ext3 defrag/file move program



I approached the problem from several different angles.

Approach #1 - Partition Must Be Unmounted to Defrag
   Advantages:
      * Minimize chance of file system corruption because
        no other processes could be modifying files.
      * Can manipulate the file system in ways that would
        otherwise be unsafe if the file system were mounted.
   Disadvantages:
      * Must directly manipulate the journal (i.e. a lot of
        tedious programming) so that the file system can
        remain consistent in the event of a power failure
      * The partition would be completely unavailable during
        the entire defrag process resulting in significant
        down time.
Ultimately rejected because of the last disadvantage.
        
Approach #2 - On-the-fly Defragmentation using the
              EXT2FS library for Block Allocation
   Advantages:
      * No down time.
      * Utilizes existing EXT3 journal programming by
        performing the defragmentation using normal
        read-write operations.
   Disadvantages:
      * Without mandatory locking, there would be no way
        to ensure that another process has not modified the
        file.
      * Because of the way the kernel buffers I/O data,
        another process could be attempting to utilize the
        same blocks as the defrag program.
Attempted, but ultimately rejected because of the last disadvantage.

Approach #3 - On-the-fly Defragmentation without a
              Block Allocation Policy
   Advantages:
      * This is a close as we can get to a "safe" on-the-fly
        defragmentation program.
      * No down time.
      * Utilizes existing EXT3 journal programming by
        performing the defragmentation using normal
        read-write operations.
   Disadvantages:
      * Without a block allocation policy, defrag would be
        nothing more than a sophisticated copy program.
      * While it is likely that the copy would be less
        fragmented than the original, without a block
        allocation policy there would be no guarantee.  The
        copy would have to be checked against the original
        before overwriting it.
      * Without mandatory locking, there would be no way
        to ensure that another process has not modified the
        file.
Attempted, but ultimately rejected because of the last disadvantage.

In order to make a defrag for EXT3 safe, you would need to do one of
two things:

1) Develop a Defrag API or some sort of File System Maintenance API
   which included Defrag support and successfully integrate it into
   the kernel.
- OR -
2) Extend Mandatory Locking to every file the system opens and closes
   and integrate such a patch into the kernel.

Either task is uphill sledding.

Joseph D. Wagner


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]