Our proudct have used squashfs for rootfs and it saves a few bucks per device. Super thanks, Squashfs! You were a perfect for us. But unfortunately, our device start to become complex so sometime we need better throughput for sequential I/O but current squashfs couldn't meet our usecase.
When I dive into the code, it has some problems. 1) Too many copy 2) Only a decompression stream buffer so that concurrent read stuck with that 3) short of readpages This patchset try to solve above problems. Just two patches are clean up so it shouldn't change any behavior. And functions factored out will be used for later patches. If they changes some behavior, it's not what I intended. :( 3rd patch removes cache usage for (not-fragemented, no-tail-end) data pages so that we can reduce memory copy. 4th patch supports multiple decompress stream buffer so concurrent read could be handled at the same time. When I tested experiment, It obviously reduces a half time. 5th patch try to implement asynchronous readahead functions so I found it can enhance about 35% with lots of I/O merging. Any comments are welcome. Thanks. Minchan Kim (5): squashfs: clean up squashfs_read_data squashfs: clean up squashfs_readpage squashfs: remove cache for normal data page squashfs: support multiple decompress stream buffer squashfs: support readpages fs/squashfs/block.c | 245 +++++++++----- fs/squashfs/cache.c | 16 +- fs/squashfs/decompressor.c | 107 +++++- fs/squashfs/decompressor.h | 27 +- fs/squashfs/file.c | 738 ++++++++++++++++++++++++++++++++++++++---- fs/squashfs/lzo_wrapper.c | 12 +- fs/squashfs/squashfs.h | 12 +- fs/squashfs/squashfs_fs_sb.h | 11 +- fs/squashfs/super.c | 44 ++- fs/squashfs/xz_wrapper.c | 20 +- fs/squashfs/zlib_wrapper.c | 12 +- 11 files changed, 1024 insertions(+), 220 deletions(-) -- 1.7.9.5 -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
