On Sun, Jan 29, 2006 at 01:53:09AM -0600, Callum Lerwick wrote: > Seems to me journaling should make it easier. Just rewind the journal. At least with ext2/3, deleting can touch a lot of data on disk, because of the way block lists are stored in inodes and indirect blocks. Block lists are large because they use an entry for every single block in the file. Other filesystems compress the lists in the case of contiguous groups of blocks (as is often the case) by using so-called extents, which store the location of the first block in the chunk and the chunk size. Journal size is a finite resource, typically 32MB, because it eats disk space and memory (it might have to be pinned by the kernel to a varying degree, depending on memory pressure, IIRC). There's only so much you could keep in there. And a single transaction can only use a portion of the total journal size. That's why ext3 fills the inode and the indirect blocks with zeroes right away. With the current design, the alternative would be a slower or less safe delete. You could try to be smarter at deleting things, but the problem with optimisations is that you also need to worry about the corner case - e.g. when the system is running out of memory - which can complicate your life a lot. -- Rudi -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
