Those benchmarks were created to checks benetifs of changing zstd implementation in https://github.com/buchgr/bazel-remote
Dataset: http://sun.aei.polsl.pl/~sdeor/corpus/silesia.zip
Results on: Intel(R) Xeon(R) Silver 4114 CPU @ 2.20GHz
Measuring CPU time (rusage utime) instead of wall time, which maybe more important in case of multiple concurrent compressions.
$ rm -rf /tmp/ramdisk/tmp && mkdir /tmp/ramdisk/tmp && go build && ./go-zstd-benchmarks -dir /tmp/ramdisk/silesia_tar -tmp_dir /tmp/ramdisk/tmp -iterations 10
Scanning files in: /tmp/ramdisk/silesia_tar
Got 1 file(s), total size: 212 MB (avg: 212 MB)
compressor inSize outSize ratio enc_time dec_time enc_speed dec_speed
IDENTITY 2.1 GB 2.1 GB 100.00 0s 0s 9.2 EB/s 9.2 EB/s
ZSTD DEFAULT 2.1 GB 675 MB 31.83 40.682418s 18.365409s 52 MB/s 115 MB/s
ZSTD BEST_SPEED 2.1 GB 735 MB 34.69 31.773793s 16.416214s 67 MB/s 129 MB/s
ZSTD CGO DEFAULT 2.1 GB 640 MB 30.19 30.840302s 3.179873s 69 MB/s 667 MB/s
ZSTD CGO SPEED 2.1 GB 736 MB 34.72 10.11669s 2.776654s 210 MB/s 763 MB/s