Re: squashfs compressed file system update?
Drew Scott Daniels wrote:
The order of the files in the initrd file can make a difference if the run
length or context of the compressor can span multiple files (large block
sizes). I don't know how one could change the order of files in an initrd
file, but I suspect it's like tar, just pass them in a different order.
This falls into the area of block allocation/placement strategy in
filesystems design. Most filesystems place blocks on disk to minimise
seek time or to minimise fragmentation when running with the 'typical'
filesystem workload. I'm not aware of any fs that places blocks to
maximise block compression ratio when compressing the entire filesystem
as a whole - though that would make sense for an initrd.
A general purpose filesystem like ext2 would probably scatter files
around the filesystem partition in clusters - each cluster will probably
contain files from the same directory or directory tree - this helps
to reduce fragmentation, and works on the assumption that files in the
same directory will be accessed together.
Romfs and squashfs compact the filesystem, and write the files to disk
in the order that the source directory is returned by readdir. Cramfs
orders the entries in each directory in alphabetical order, as this
potentially speeds up directory lookup. Because directories and the
inode table are the same thing in cramfs, this means the inodes in each
directory will be in alphabetical order, and so the file data is likely
to be written likewise. Of course for cramfs and squashfs this is not
much of an issue because they're already compressed.
The order of files would make no difference if files are compressed
independently like squashfs does.
You're right.
Phillip Lougher:
I'm also not subscribed to debian-boot thus I can't get the threading
right, but you did. I read the mailing list at
http://lists.debian.org/debian-boot and several others on
lists.debian.org.
Would you be willing to try to get squashfs to be made part of the
standard Debian Linux kernel? I'm not sure how, or what the reaction would
be.
I'd like to get squashfs accepted as part of the Debian Linux kernel,
though I daresay it's completely out of my control.
I've submitted a request for packaging which you can see at
http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=179672 It's likely that
the Debian Linux kernel maintainers would not include squashfs, but would
suggest that it be made a patch package. If the boot system team finds it
useful enough then maybe that would be enough to have squashfs include in
the Debian Linux kernel.
That would be good - if anyone else reads this who is involved in this
process, then i'm happy to help.
Phillip
Drew Daniels
Reply to: