We've seen that overall, LZ4 is a better compression algorithm than LZJB. It compresses and decompresses faster, and gets a better compression ratio. It's been in use long enough for us to shake out the few initial problems with the implementation.
I think we should change the default compression algorithm in ZFS to LZ4. Specifically, if the LZ4 feature is enabled (e.g. via "zpool set feature@lz4_compress=enabled"), and you do "zfs set compression=on", then it will compress with LZ4 rather than LZJB. We should probably also compress metadata with LZ4 if the feature is enabled (after verifying that indirect blocks compress at least as well as LZ4). I am pretty busy with other ZFS work at the moment, but I'd be happy to guide / mentor someone from the community who would like to implement this. It would also make a good hackathon project at the OpenZFS Developer Summit <http://www.open-zfs.org/wiki/OpenZFS_Developer_Summit_2014> (November 2014 in San Francisco). --matt
_______________________________________________ developer mailing list [email protected] http://lists.open-zfs.org/mailman/listinfo/developer
