Skip to content

WSS Local Performance

Measurements come from independent local Release loopback processes, not CI.

Metric definitions and methodology.

Average process CPU uses 100% for one logical CPU and can exceed 100% with multiple threads; estimated values are labeled, and incomplete sets of three CPU samples show N/A. On this 14-logical-CPU host, machine share = process CPU% / 14; 100% of one CPU is 7.14% of the machine.

No added computation

Throughput

p99 RTT

All Measured Profiles

64 B / 1 clients / window=1

Dependency / model Round trips/s MiB/s p50 / us p95 / us p99 / us CPU / % RSS / MiB
Boost.Asio / 2 contexts / 2 threads 46,068 [45,672, 47,126] 5.62 20.38 26.50 35.83 98.5 (estimated) 14.89
Boost.Asio / 4 contexts / 4 threads 45,865 [45,500, 46,446] 5.60 20.50 26.08 34.92 98.8 (estimated) 14.92
Boost.Asio / 1 context / 1 thread 49,119 [48,765, 49,179] 6.00 19.29 27.38 35.21 88.7 (estimated) 14.84
Boost.Asio / 1 context / 2 threads 44,583 [43,831, 46,348] 5.44 21.38 28.21 40.62 129.0 (estimated) 14.95
Boost.Asio / 1 context / 4 threads 32,948 [32,662, 33,585] 4.02 28.29 38.75 55.33 158.5 (estimated) 15.09
Standalone Asio / 2 contexts / 2 threads 47,357 [47,066, 47,441] 5.78 20.04 25.46 35.62 101.0 (estimated) 15.17
Standalone Asio / 4 contexts / 4 threads 46,759 [46,535, 47,104] 5.71 20.29 25.67 33.71 101.1 (estimated) 15.34
Standalone Asio / 1 context / 1 thread 49,871 [49,348, 50,287] 6.09 19.04 28.58 34.04 89.7 (estimated) 15.19
Standalone Asio / 1 context / 2 threads 44,429 [44,279, 44,673] 5.42 21.46 28.33 39.88 130.6 (estimated) 15.27
Standalone Asio / 1 context / 4 threads 32,617 [32,452, 33,063] 3.98 28.71 37.67 59.75 150.5 (estimated) 15.22

1024 B / 1 clients / window=1

Dependency / model Round trips/s MiB/s p50 / us p95 / us p99 / us CPU / % RSS / MiB
Boost.Asio / 2 contexts / 2 threads 40,311 [40,196, 40,665] 78.73 23.58 27.88 35.79 101.1 (estimated) 14.94
Boost.Asio / 4 contexts / 4 threads 39,829 [39,757, 40,695] 77.79 23.62 28.54 35.42 100.5 (estimated) 14.92
Boost.Asio / 1 context / 1 thread 46,789 [46,732, 47,103] 91.38 19.83 27.21 37.29 88.6 (estimated) 14.91
Boost.Asio / 1 context / 2 threads 38,014 [38,014, 39,298] 74.25 24.46 30.75 40.46 126.0 (estimated) 14.95
Boost.Asio / 1 context / 4 threads 30,500 [30,457, 31,791] 59.57 30.54 38.71 56.50 147.6 (estimated) 14.89
Standalone Asio / 2 contexts / 2 threads 43,011 [42,901, 43,635] 84.01 21.42 28.04 36.12 101.8 (estimated) 15.27
Standalone Asio / 4 contexts / 4 threads 43,070 [42,382, 43,278] 84.12 21.33 28.17 36.42 101.6 (estimated) 15.17
Standalone Asio / 1 context / 1 thread 46,985 [46,883, 48,096] 91.77 20.33 26.17 35.08 90.1 (estimated) 15.08
Standalone Asio / 1 context / 2 threads 38,191 [36,820, 38,211] 74.59 24.54 31.21 43.62 125.8 (estimated) 15.30
Standalone Asio / 1 context / 4 threads 30,354 [30,325, 30,490] 59.29 30.92 38.21 58.12 150.2 (estimated) 15.39

16384 B / 1 clients / window=1

Dependency / model Round trips/s MiB/s p50 / us p95 / us p99 / us CPU / % RSS / MiB
Boost.Asio / 2 contexts / 2 threads 5,547 [5,547, 5,581] 173.36 171.25 192.58 216.83 75.7 (estimated) 15.09
Boost.Asio / 4 contexts / 4 threads 5,552 [5,407, 5,567] 173.51 172.46 193.83 220.96 76.1 (estimated) 15.05
Boost.Asio / 1 context / 1 thread 6,291 [6,104, 6,331] 196.58 147.25 165.92 185.88 61.8 (estimated) 14.94
Boost.Asio / 1 context / 2 threads 12,135 [12,058, 12,411] 379.21 70.42 86.08 122.71 147.0 (estimated) 15.17
Boost.Asio / 1 context / 4 threads 8,612 [8,458, 8,645] 269.13 102.08 136.88 174.62 214.4 (estimated) 15.25
Standalone Asio / 2 contexts / 2 threads 5,538 [5,530, 5,553] 173.07 168.58 195.04 233.92 77.6 (estimated) 15.47
Standalone Asio / 4 contexts / 4 threads 5,554 [5,552, 5,610] 173.55 168.42 192.58 225.25 77.4 (estimated) 15.34
Standalone Asio / 1 context / 1 thread 6,074 [6,063, 6,096] 189.82 152.54 172.67 196.42 61.5 (estimated) 15.27
Standalone Asio / 1 context / 2 threads 11,615 [11,498, 11,634] 362.96 74.46 88.42 125.46 146.2 (estimated) 15.42
Standalone Asio / 1 context / 4 threads 8,788 [8,384, 8,858] 274.62 98.42 135.50 177.21 210.8 (estimated) 15.44

64 B / 16 clients / window=16

Dependency / model Round trips/s MiB/s p50 / us p95 / us p99 / us CPU / % RSS / MiB
Boost.Asio / 2 contexts / 2 threads 192,250 [175,990, 200,495] 23.47 1,254.75 1,993.62 2,337.58 198.4 (estimated) 35.55
Boost.Asio / 4 contexts / 4 threads 252,537 [241,855, 267,290] 30.83 1,012.88 1,243.50 1,286.79 395.1 (estimated) 36.02
Boost.Asio / 1 context / 1 thread 100,069 [99,502, 101,709] 12.22 2,536.08 2,715.96 2,902.75 97.4 (estimated) 35.42
Boost.Asio / 1 context / 2 threads 154,293 [138,988, 154,559] 18.83 1,642.75 1,901.88 2,090.46 193.8 (estimated) 35.61
Boost.Asio / 1 context / 4 threads 170,278 [165,546, 172,461] 20.79 1,441.46 1,920.38 2,254.38 363.6 (estimated) 35.83
Standalone Asio / 2 contexts / 2 threads 172,705 [165,864, 173,198] 21.08 1,376.21 2,267.50 2,512.67 198.6 (estimated) 36.03
Standalone Asio / 4 contexts / 4 threads 242,999 [241,687, 245,351] 29.66 1,053.42 1,248.38 1,294.75 396.4 (estimated) 36.42
Standalone Asio / 1 context / 1 thread 95,753 [95,049, 95,843] 11.69 2,650.50 2,832.04 3,015.54 99.7 (estimated) 35.98
Standalone Asio / 1 context / 2 threads 153,083 [152,977, 154,594] 18.69 1,603.50 2,115.58 2,448.08 197.7 (estimated) 36.03
Standalone Asio / 1 context / 4 threads 192,089 [189,945, 198,778] 23.45 1,234.54 1,754.83 1,904.42 387.2 (estimated) 36.34

1024 B / 16 clients / window=16

Dependency / model Round trips/s MiB/s p50 / us p95 / us p99 / us CPU / % RSS / MiB
Boost.Asio / 2 contexts / 2 threads 145,195 [143,721, 167,918] 283.58 1,550.00 2,672.92 2,872.33 199.0 (estimated) 36.39
Boost.Asio / 4 contexts / 4 threads 225,102 [208,872, 228,893] 439.65 1,125.71 1,494.21 1,553.58 397.2 (estimated) 36.89
Boost.Asio / 1 context / 1 thread 82,094 [81,933, 84,959] 160.34 3,097.38 3,278.00 3,407.67 94.6 (estimated) 36.14
Boost.Asio / 1 context / 2 threads 135,355 [124,418, 135,942] 264.36 1,864.33 2,213.62 2,462.00 195.4 (estimated) 36.48
Boost.Asio / 1 context / 4 threads 160,568 [157,002, 161,967] 313.61 1,504.33 2,046.54 2,235.21 369.7 (estimated) 36.80
Standalone Asio / 2 contexts / 2 threads 144,290 [142,592, 145,176] 281.82 1,635.46 2,736.67 2,984.75 198.5 (estimated) 36.70
Standalone Asio / 4 contexts / 4 threads 214,228 [213,545, 217,103] 418.41 1,172.12 1,447.12 1,491.67 397.5 (estimated) 42.11
Standalone Asio / 1 context / 1 thread 77,864 [76,652, 78,341] 152.08 3,260.96 3,475.08 3,739.62 99.5 (estimated) 36.55
Standalone Asio / 1 context / 2 threads 133,747 [132,771, 134,695] 261.22 1,840.50 2,367.12 2,804.96 197.9 (estimated) 36.66
Standalone Asio / 1 context / 4 threads 177,998 [155,022, 179,441] 347.65 1,345.04 1,885.54 2,017.00 389.7 (estimated) 37.97

16384 B / 16 clients / window=16

Dependency / model Round trips/s MiB/s p50 / us p95 / us p99 / us CPU / % RSS / MiB
Boost.Asio / 2 contexts / 2 threads 27,428 [23,592, 29,601] 857.12 9,153.67 10,818.46 11,706.54 197.9 (estimated) 48.28
Boost.Asio / 4 contexts / 4 threads 43,047 [42,524, 43,361] 1,345.22 5,869.62 6,714.21 7,012.58 384.9 (estimated) 62.72
Boost.Asio / 1 context / 1 thread 14,269 [14,073, 14,360] 445.91 17,800.46 18,697.88 19,819.21 96.6 (estimated) 48.22
Boost.Asio / 1 context / 2 threads 24,178 [23,704, 24,545] 755.56 10,368.79 11,862.96 12,620.54 196.0 (estimated) 48.05
Boost.Asio / 1 context / 4 threads 29,649 [28,303, 31,402] 926.53 8,441.38 9,689.67 10,220.75 367.7 (estimated) 48.17
Standalone Asio / 2 contexts / 2 threads 26,700 [26,196, 26,831] 834.37 9,170.42 11,300.71 12,100.29 199.0 (estimated) 48.09
Standalone Asio / 4 contexts / 4 threads 41,250 [40,847, 42,032] 1,289.06 6,226.42 6,766.92 6,979.54 390.4 (estimated) 48.88
Standalone Asio / 1 context / 1 thread 13,795 [13,688, 13,837] 431.11 18,341.62 19,153.75 21,286.96 98.0 (estimated) 47.88
Standalone Asio / 1 context / 2 threads 24,551 [23,839, 24,614] 767.23 10,401.75 11,389.29 11,687.58 197.7 (estimated) 48.48
Standalone Asio / 1 context / 4 threads 31,271 [30,987, 31,342] 977.22 8,056.33 9,162.50 9,669.79 383.9 (estimated) 48.58

Tables show repeated-run medians; throughput brackets show min/max, not confidence intervals. Latencies are medians of per-run sampled percentiles, without pooling samples.

Navigation

Type to search…

↑↓ navigate↵ selectEsc close