> On Nov 1, 2018, at 5:57 AM, Timofey Titovets <nefelim...@gmail.com> wrote: > > ср, 31 окт. 2018 г. в 21:12, Nick Terrell <terre...@fb.com>: >> >> From: Jennifer Liu <jenniferliu...@fb.com> >> >> Adds zstd compression level support to btrfs. Zstd requires >> different amounts of memory for each level, so the design had >> to be modified to allow set_level() to allocate memory. We >> preallocate one workspace of the maximum size to guarantee >> forward progress. This feature is expected to be useful for >> read-mostly filesystems, or when creating images. >> >> Benchmarks run in qemu on Intel x86 with a single core. >> The benchmark measures the time to copy the Silesia corpus [0] to >> a btrfs filesystem 10 times, then read it back. >> >> The two important things to note are: >> - The decompression speed and memory remains constant. >> The memory required to decompress is the same as level 1. >> - The compression speed and ratio will vary based on the source. >> >> Level Ratio Compression Decompression Compression Memory >> 1 2.59 153 MB/s 112 MB/s 0.8 MB >> 2 2.67 136 MB/s 113 MB/s 1.0 MB >> 3 2.72 106 MB/s 115 MB/s 1.3 MB >> 4 2.78 86 MB/s 109 MB/s 0.9 MB >> 5 2.83 69 MB/s 109 MB/s 1.4 MB >> 6 2.89 53 MB/s 110 MB/s 1.5 MB >> 7 2.91 40 MB/s 112 MB/s 1.4 MB >> 8 2.92 34 MB/s 110 MB/s 1.8 MB >> 9 2.93 27 MB/s 109 MB/s 1.8 MB >> 10 2.94 22 MB/s 109 MB/s 1.8 MB >> 11 2.95 17 MB/s 114 MB/s 1.8 MB >> 12 2.95 13 MB/s 113 MB/s 1.8 MB >> 13 2.95 10 MB/s 111 MB/s 2.3 MB >> 14 2.99 7 MB/s 110 MB/s 2.6 MB >> 15 3.03 6 MB/s 110 MB/s 2.6 MB >> >> [0] http://sun.aei.polsl.pl/~sdeor/index.php?page=silesia >> >> Signed-off-by: Jennifer Liu <jenniferliu...@fb.com> >> Signed-off-by: Nick Terrell <terre...@fb.com> >> Reviewed-by: Omar Sandoval <osan...@fb.com>
<snip> > > You didn't mention, so: > Did you test compression ratio/performance with compress-force or just > compress? I tested with compress-force, since I just reused my script from before [0]. I reran some levels with compress and got these numbers: Level Ratio Compression Decompression 1 2.21 158 MB/s 113 MB/s 3 2.28 117 MB/s 113 MB/s 5 2.32 81 MB/s 112 MB/s 7 2.37 47 MB/s 116 MB/s 15 2.41 7 MB/s 115 MB/s Using compress probably makes sense with lower levels, to get higher write speeds, but if you're using higher compression levels, you'll likely want to use compress-force, since you likely don't care too much about write speeds. Obviously this will depend on the data you're compressing. [0] https://gist.github.com/terrelln/51ed3c9da6f94d613c01fcdae60567e8 -Nick > > Thanks. > > -- > Have a nice day, > Timofey.
