On Thu, May 24, 2007 at 08:15:28PM +0900, Kazuki Mizushima wrote: > >>Now, I investigate sparse files copying. > > > >There's no easy way to detect if a file is sparse - particularly in python. > >I think the best option is to detect sequences of zeros. eg, if you are > >reading the source file 4096 bytes at a time, check to see if that 4096 > >set of bytes are all zeros. If they are, then seek() forward 4096 bytes > >instead of writing the set of zeros. > > When copying a sparse file, I change read/write buffer to 4096 because it > is 4096 in cp.c. And I add the "--nonsparse" option as virt-install. > Default behavior is creating a sparse. Basically when destination file/disk > doesn't > exist, create a sparse file and do sparselly copying (means lseek when > detecting > 4096 sequences of zeros) to that sparse file. When destination file exists, > do non-sparselly copying(means no lseek when detecting 4096 sequences of > zeros) > to the existing file. > > In the above-mentioned, as a result of measuring the performance of copying > a sparse, I get the performance equivalent to the command of 'cp > --sparse=always'. Thanks, I have comitted this patch. Dan. -- |=- Red Hat, Engineering, Emerging Technologies, Boston. +1 978 392 2496 -=| |=- Perl modules: http://search.cpan.org/~danberr/ -=| |=- Projects: http://freshmeat.net/~danielpb/ -=| |=- GnuPG: 7D3B9505 F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505 -=|
