Remove rand_chacha

[dhardy]

Summary

Source moved: rust-random/rngs#68

[dhardy]

A quick benchmark run shows no very significant changes (the bar of "significant" being fairly high due to unpinned CPU freq.).

Benchmark

$ cargo bench -- chacha
   Compiling rand_pcg v0.9.0 (/home/dhardy/projects/rand/rand/rand_pcg)
   Compiling rand v0.9.2 (/home/dhardy/projects/rand/rand)
   Compiling benches v0.1.0 (/home/dhardy/projects/rand/rand/benches)
    Finished `bench` profile [optimized] target(s) in 5.97s
     Running benches/array.rs (/home/dhardy-extra/.cache/cargo-build/release/deps/array-af84aee0fd253b08)
     Running benches/bool.rs (/home/dhardy-extra/.cache/cargo-build/release/deps/bool-0b6ead89b6d1c550)
     Running benches/generators.rs (/home/dhardy-extra/.cache/cargo-build/release/deps/generators-827c3b4b67548787)
random_bytes/chacha8    time:   [245.77 ns 246.39 ns 246.93 ns]
                        thrpt:  [3.8622 GiB/s 3.8706 GiB/s 3.8804 GiB/s]
                 change:
                        time:   [-0.2241% +0.6495% +1.5004%] (p = 0.14 > 0.05)
                        thrpt:  [-1.4782% -0.6453% +0.2246%]
                        No change in performance detected.
Found 2 outliers among 100 measurements (2.00%)
  1 (1.00%) high mild
  1 (1.00%) high severe
random_bytes/chacha12   time:   [332.48 ns 332.90 ns 333.31 ns]
                        thrpt:  [2.8612 GiB/s 2.8648 GiB/s 2.8683 GiB/s]
                 change:
                        time:   [+8.0156% +8.6076% +9.1503%] (p = 0.00 < 0.05)
                        thrpt:  [-8.3832% -7.9254% -7.4208%]
                        Performance has regressed.
Found 9 outliers among 100 measurements (9.00%)
  3 (3.00%) low mild
  3 (3.00%) high mild
  3 (3.00%) high severe
random_bytes/chacha20   time:   [444.71 ns 446.35 ns 448.80 ns]
                        thrpt:  [2.1250 GiB/s 2.1366 GiB/s 2.1445 GiB/s]
                 change:
                        time:   [-1.7830% -1.1872% -0.5086%] (p = 0.00 < 0.05)
                        thrpt:  [+0.5112% +1.2015% +1.8154%]
                        Change within noise threshold.
Found 11 outliers among 100 measurements (11.00%)
  2 (2.00%) low mild
  5 (5.00%) high mild
  4 (4.00%) high severe

random_u32/chacha8      time:   [957.66 ps 960.68 ps 963.89 ps]
                        thrpt:  [3.8649 GiB/s 3.8778 GiB/s 3.8900 GiB/s]
                 change:
                        time:   [+1.4652% +1.9002% +2.3720%] (p = 0.00 < 0.05)
                        thrpt:  [-2.3170% -1.8648% -1.4441%]
                        Performance has regressed.
Found 130 outliers among 1000 measurements (13.00%)
  22 (2.20%) high mild
  108 (10.80%) high severe
random_u32/chacha12     time:   [1.2200 ns 1.2247 ns 1.2297 ns]
                        thrpt:  [3.0295 GiB/s 3.0417 GiB/s 3.0534 GiB/s]
                 change:
                        time:   [-2.5279% -2.0627% -1.5850%] (p = 0.00 < 0.05)
                        thrpt:  [+1.6106% +2.1062% +2.5934%]
                        Performance has improved.
Found 148 outliers among 1000 measurements (14.80%)
  67 (6.70%) high mild
  81 (8.10%) high severe
random_u32/chacha20     time:   [1.6805 ns 1.6823 ns 1.6842 ns]
                        thrpt:  [2.2119 GiB/s 2.2144 GiB/s 2.2168 GiB/s]
                 change:
                        time:   [-5.2598% -4.9730% -4.6967%] (p = 0.00 < 0.05)
                        thrpt:  [+4.9281% +5.2333% +5.5518%]
                        Performance has improved.
Found 59 outliers among 1000 measurements (5.90%)
  23 (2.30%) high mild
  36 (3.60%) high severe

random_u64/chacha8      time:   [1.3738 ns 1.3767 ns 1.3798 ns]
                        thrpt:  [5.3998 GiB/s 5.4119 GiB/s 5.4234 GiB/s]
                 change:
                        time:   [-5.2813% -4.8917% -4.5020%] (p = 0.00 < 0.05)
                        thrpt:  [+4.7143% +5.1433% +5.5758%]
                        Performance has improved.
Found 118 outliers among 1000 measurements (11.80%)
  4 (0.40%) low mild
  49 (4.90%) high mild
  65 (6.50%) high severe
random_u64/chacha12     time:   [1.9632 ns 1.9750 ns 1.9872 ns]
                        thrpt:  [3.7494 GiB/s 3.7725 GiB/s 3.7951 GiB/s]
                 change:
                        time:   [-2.9714% -2.3478% -1.7632%] (p = 0.00 < 0.05)
                        thrpt:  [+1.7949% +2.4042% +3.0624%]
                        Performance has improved.
Found 130 outliers among 1000 measurements (13.00%)
  1 (0.10%) low mild
  19 (1.90%) high mild
  110 (11.00%) high severe
random_u64/chacha20     time:   [2.8760 ns 2.8866 ns 2.8979 ns]
                        thrpt:  [2.5711 GiB/s 2.5811 GiB/s 2.5906 GiB/s]
                 change:
                        time:   [-8.6350% -8.1851% -7.7253%] (p = 0.00 < 0.05)
                        thrpt:  [+8.3720% +8.9148% +9.4511%]
                        Performance has improved.
Found 140 outliers among 1000 measurements (14.00%)
  45 (4.50%) high mild
  95 (9.50%) high severe

init_gen/chacha8        time:   [32.132 ns 32.186 ns 32.247 ns]
                        change: [+0.6341% +1.2175% +1.8555%] (p = 0.00 < 0.05)
                        Change within noise threshold.
Found 5 outliers among 100 measurements (5.00%)
  3 (3.00%) high mild
  2 (2.00%) high severe
init_gen/chacha12       time:   [31.814 ns 32.010 ns 32.260 ns]
                        change: [-0.8760% +0.0020% +0.7705%] (p = 1.00 > 0.05)
                        No change in performance detected.
Found 11 outliers among 100 measurements (11.00%)
  4 (4.00%) high mild
  7 (7.00%) high severe
init_gen/chacha20       time:   [32.064 ns 32.110 ns 32.161 ns]
                        change: [-3.6865% -2.5900% -1.5765%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 10 outliers among 100 measurements (10.00%)
  5 (5.00%) low mild
  2 (2.00%) high mild
  3 (3.00%) high severe

init_from_u64/chacha8   time:   [41.519 ns 41.628 ns 41.731 ns]
                        change: [-4.8846% -4.5025% -4.0995%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild
init_from_u64/chacha12  time:   [41.760 ns 41.833 ns 41.911 ns]
                        change: [-5.0056% -4.5781% -4.1155%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  1 (1.00%) high mild
  2 (2.00%) high severe
init_from_u64/chacha20  time:   [40.997 ns 41.045 ns 41.101 ns]
                        change: [-9.1845% -8.4379% -7.7492%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 6 outliers among 100 measurements (6.00%)
  6 (6.00%) high mild

init_from_seed/chacha8  time:   [32.865 ns 32.930 ns 33.009 ns]
                        change: [+7.3252% +7.8271% +8.3239%] (p = 0.00 < 0.05)
                        Performance has regressed.
init_from_seed/chacha12 time:   [33.311 ns 33.357 ns 33.401 ns]
                        change: [+6.7469% +7.5707% +8.3332%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 6 outliers among 100 measurements (6.00%)
  3 (3.00%) low mild
  2 (2.00%) high mild
  1 (1.00%) high severe
init_from_seed/chacha20 time:   [33.337 ns 33.615 ns 33.924 ns]
                        change: [+6.3530% +7.5427% +8.6519%] (p = 0.00 < 0.05)
                        Performance has regressed.
Found 8 outliers among 100 measurements (8.00%)
  3 (3.00%) high mild
  5 (5.00%) high severe

reseeding_bytes/chacha20_4k
                        time:   [395.08 µs 396.10 µs 397.30 µs]
                        thrpt:  [2.4580 GiB/s 2.4654 GiB/s 2.4718 GiB/s]
                 change:
                        time:   [-5.1210% -4.2470% -3.3999%] (p = 0.00 < 0.05)
                        thrpt:  [+3.5195% +4.4353% +5.3975%]
                        Performance has improved.
Found 19 outliers among 100 measurements (19.00%)
  3 (3.00%) low severe
  7 (7.00%) low mild
  3 (3.00%) high mild
  6 (6.00%) high severe
reseeding_bytes/chacha20_16k
                        time:   [380.01 µs 380.82 µs 381.78 µs]
                        thrpt:  [2.5579 GiB/s 2.5644 GiB/s 2.5698 GiB/s]
                 change:
                        time:   [-0.3435% +0.0662% +0.5591%] (p = 0.77 > 0.05)
                        thrpt:  [-0.5560% -0.0661% +0.3447%]
                        No change in performance detected.
Found 12 outliers among 100 measurements (12.00%)
  1 (1.00%) low mild
  6 (6.00%) high mild
  5 (5.00%) high severe
reseeding_bytes/chacha20_32k
                        time:   [377.20 µs 377.86 µs 378.63 µs]
                        thrpt:  [2.5792 GiB/s 2.5844 GiB/s 2.5890 GiB/s]
                 change:
                        time:   [-1.9226% -1.2419% -0.4789%] (p = 0.00 < 0.05)
                        thrpt:  [+0.4812% +1.2575% +1.9603%]
                        Change within noise threshold.
Found 23 outliers among 100 measurements (23.00%)
  1 (1.00%) low severe
  1 (1.00%) low mild
  9 (9.00%) high mild
  12 (12.00%) high severe
reseeding_bytes/chacha20_64k
                        time:   [377.04 µs 378.48 µs 379.98 µs]
                        thrpt:  [2.5700 GiB/s 2.5803 GiB/s 2.5901 GiB/s]
                 change:
                        time:   [-0.3950% +0.1281% +0.6287%] (p = 0.62 > 0.05)
                        thrpt:  [-0.6248% -0.1279% +0.3965%]
                        No change in performance detected.
Found 12 outliers among 100 measurements (12.00%)
  8 (8.00%) high mild
  4 (4.00%) high severe
reseeding_bytes/chacha20_256k
                        time:   [373.82 µs 375.39 µs 376.93 µs]
                        thrpt:  [2.5908 GiB/s 2.6014 GiB/s 2.6124 GiB/s]
                 change:
                        time:   [-2.3042% -1.5835% -0.7933%] (p = 0.00 < 0.05)
                        thrpt:  [+0.7996% +1.6090% +2.3585%]
                        Change within noise threshold.
Found 13 outliers among 100 measurements (13.00%)
  8 (8.00%) high mild
  5 (5.00%) high severe
reseeding_bytes/chacha20_1024k
                        time:   [376.31 µs 377.46 µs 378.89 µs]
                        thrpt:  [2.5774 GiB/s 2.5872 GiB/s 2.5951 GiB/s]
                 change:
                        time:   [-1.3651% -0.8050% -0.2695%] (p = 0.00 < 0.05)
                        thrpt:  [+0.2702% +0.8116% +1.3840%]
                        Change within noise threshold.
Found 2 outliers among 100 measurements (2.00%)
  2 (2.00%) high mild

[dhardy]

It may make sense to hold off merging this for a while. It conflicts with #1659.

[newpavlov]

5-10% of performance drop is a bit too much for noise even with unpinned frequency considering that criterion does a proper CPU warm-up. I will try to take a closer look at the RNG implementation later.

[dhardy]

Time

0.65%
8.61%
−1.19%
1.90%
−2.06%
−4.97%
−4.89%
−2.35%
−8.19%
1.22%
0.00%
−2.59%
−4.50%
−4.58%
−8.44%
7.83%
7.54%
−4.25%
0.07%
0.13%
−1.58%
−0.80%
Average
−1.02%

[dhardy]

@hpenne reported the new ChaCha impls have doubled perf. on Apple M1.

[dhardy]

Updated.

We've been depending on chacha20 for a while now; I no longer see a reason to maintain rand_chacha.

[vks]

Before removing it, should we make a final release updating the README to say it's deprecated?

[dhardy]

Should we mark it as deprecated? Certainly not now (no final release of chacha20 yet).

Later? I don't see much point since it won't be compatible with the next rand_core version. Also, I don't see any need to force people to migrate (no known security issues or other major problems).

So I feel that we shouldn't.

[tarcieri]

So it's just being migrated to a different repo then, right?

[newpavlov]

I think we can move rand_chacha to the rngs repo and release v0.10 with the new rand_core. Next, after chacha20 is released and assuming there are no problems with it, we can release empty rand_chacha v0.11 with suggestion to migrate to chacha20, thus effectively deprecating rand_chacha. Those who track outdated dependencies will migrate, those who don't care will stay on v0.10.