Skip to content

TCP+TLS 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 51,540 [50,864, 51,618] 6.29 18.83 23.54 27.58 88.4 (estimated) 13.48
Boost.Asio / 4 contexts / 4 threads 51,150 [50,499, 51,535] 6.24 18.92 23.83 28.25 88.8 (estimated) 13.61
Boost.Asio / 1 context / 1 thread 49,767 [49,522, 50,002] 6.08 20.50 23.96 27.75 83.2 (estimated) 13.34
Boost.Asio / 1 context / 2 threads 46,177 [44,566, 46,698] 5.64 21.08 25.79 32.21 116.3 (estimated) 13.56
Boost.Asio / 1 context / 4 threads 39,558 [38,518, 39,641] 4.83 24.67 29.88 35.17 152.3 (estimated) 13.48
Standalone Asio / 2 contexts / 2 threads 46,297 [46,221, 50,170] 5.65 20.29 26.75 37.75 90.6 (estimated) 13.88
Standalone Asio / 4 contexts / 4 threads 47,580 [46,594, 51,310] 5.81 20.12 26.04 31.79 90.2 (estimated) 13.97
Standalone Asio / 1 context / 1 thread 48,332 [47,075, 49,726] 5.90 20.46 24.96 29.38 81.9 (estimated) 13.80
Standalone Asio / 1 context / 2 threads 43,561 [42,611, 45,615] 5.32 22.75 26.92 31.83 117.5 (estimated) 13.86
Standalone Asio / 1 context / 4 threads 37,406 [37,152, 38,608] 4.57 25.17 35.29 54.33 149.1 (estimated) 13.89

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 46,758 [44,557, 47,324] 91.32 20.00 25.50 29.21 94.0 (estimated) 13.58
Boost.Asio / 4 contexts / 4 threads 46,710 [44,717, 46,936] 91.23 20.04 25.42 29.29 94.0 (estimated) 13.53
Boost.Asio / 1 context / 1 thread 47,572 [47,236, 47,724] 92.91 20.08 24.62 28.29 86.2 (estimated) 13.48
Boost.Asio / 1 context / 2 threads 43,103 [42,811, 43,395] 84.19 22.12 26.71 32.00 120.9 (estimated) 13.50
Boost.Asio / 1 context / 4 threads 36,724 [36,587, 36,965] 71.73 25.83 30.83 37.04 137.7 (estimated) 13.59
Standalone Asio / 2 contexts / 2 threads 43,676 [43,206, 46,163] 85.31 21.42 27.12 32.79 96.8 (estimated) 13.94
Standalone Asio / 4 contexts / 4 threads 44,001 [41,953, 45,610] 85.94 21.00 27.42 33.79 95.4 (estimated) 13.95
Standalone Asio / 1 context / 1 thread 47,247 [46,643, 47,259] 92.28 20.38 25.12 29.67 85.4 (estimated) 13.84
Standalone Asio / 1 context / 2 threads 40,759 [40,652, 41,739] 79.61 23.38 27.92 32.88 121.7 (estimated) 13.83
Standalone Asio / 1 context / 4 threads 35,933 [34,939, 35,966] 70.18 26.33 32.58 44.83 137.5 (estimated) 13.94

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 12,852 [12,311, 12,917] 401.63 64.42 77.08 99.25 89.6 (estimated) 13.66
Boost.Asio / 4 contexts / 4 threads 12,671 [12,503, 12,696] 395.97 65.21 79.83 105.50 89.8 (estimated) 13.69
Boost.Asio / 1 context / 1 thread 12,447 [12,345, 12,548] 388.97 70.33 79.46 88.17 67.7 (estimated) 13.64
Boost.Asio / 1 context / 2 threads 17,709 [17,707, 17,942] 553.42 44.92 52.42 70.12 121.9 (estimated) 13.70
Boost.Asio / 1 context / 4 threads 17,121 [17,050, 17,865] 535.03 46.38 56.25 77.62 129.0 (estimated) 13.72
Standalone Asio / 2 contexts / 2 threads 12,901 [12,046, 12,955] 403.16 65.25 75.29 81.50 89.6 (estimated) 14.06
Standalone Asio / 4 contexts / 4 threads 12,801 [12,792, 12,835] 400.05 65.04 77.00 88.96 89.4 (estimated) 14.06
Standalone Asio / 1 context / 1 thread 12,426 [12,330, 13,034] 388.31 69.21 79.83 89.17 67.3 (estimated) 13.98
Standalone Asio / 1 context / 2 threads 17,193 [17,033, 17,516] 537.30 45.50 57.42 107.75 121.0 (estimated) 14.03
Standalone Asio / 1 context / 4 threads 17,102 [15,336, 17,141] 534.42 47.00 55.00 77.79 128.7 (estimated) 14.14

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 235,367 [196,640, 245,898] 28.73 1,018.79 1,519.50 1,998.17 198.4 (estimated) 33.94
Boost.Asio / 4 contexts / 4 threads 295,973 [288,969, 296,431] 36.13 877.04 1,184.17 1,242.50 392.6 (estimated) 34.47
Boost.Asio / 1 context / 1 thread 130,701 [128,800, 135,530] 15.95 1,963.08 2,056.96 2,249.08 98.6 (estimated) 33.75
Boost.Asio / 1 context / 2 threads 191,203 [188,374, 196,004] 23.34 1,317.83 1,491.96 1,626.58 193.2 (estimated) 33.94
Boost.Asio / 1 context / 4 threads 234,844 [231,059, 246,525] 28.67 1,058.29 1,447.46 1,674.04 370.8 (estimated) 34.12
Standalone Asio / 2 contexts / 2 threads 247,980 [231,956, 250,996] 30.27 985.17 1,310.92 1,797.50 199.1 (estimated) 34.41
Standalone Asio / 4 contexts / 4 threads 302,330 [268,790, 315,770] 36.91 836.04 1,160.88 1,232.54 392.6 (estimated) 34.77
Standalone Asio / 1 context / 1 thread 131,202 [129,019, 145,120] 16.02 1,936.08 2,059.71 2,158.42 97.0 (estimated) 34.12
Standalone Asio / 1 context / 2 threads 184,158 [180,633, 198,705] 22.48 1,301.88 1,471.75 1,759.92 195.9 (estimated) 34.28
Standalone Asio / 1 context / 4 threads 247,099 [233,439, 254,342] 30.16 1,033.75 1,295.92 1,493.46 372.9 (estimated) 34.52

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 188,519 [185,430, 201,912] 368.20 1,246.83 2,019.71 2,401.83 198.5 (estimated) 34.81
Boost.Asio / 4 contexts / 4 threads 259,126 [251,591, 278,153] 506.11 996.62 1,240.38 1,305.29 394.7 (estimated) 35.53
Boost.Asio / 1 context / 1 thread 102,232 [100,722, 103,783] 199.67 2,471.08 2,641.67 3,407.38 95.8 (estimated) 34.58
Boost.Asio / 1 context / 2 threads 169,689 [166,315, 170,448] 331.42 1,505.50 1,710.42 1,916.29 195.5 (estimated) 34.81
Boost.Asio / 1 context / 4 threads 230,046 [215,719, 230,924] 449.31 1,115.12 1,332.04 1,529.29 378.6 (estimated) 35.16
Standalone Asio / 2 contexts / 2 threads 188,619 [181,678, 197,707] 368.40 1,249.08 2,073.33 2,346.62 198.4 (estimated) 35.06
Standalone Asio / 4 contexts / 4 threads 248,090 [247,535, 267,255] 484.55 1,035.12 1,257.46 1,325.46 394.6 (estimated) 35.42
Standalone Asio / 1 context / 1 thread 105,254 [104,778, 108,683] 205.57 2,416.46 2,495.38 2,657.79 96.1 (estimated) 34.88
Standalone Asio / 1 context / 2 threads 168,696 [161,784, 171,526] 329.48 1,504.12 1,802.08 2,020.21 196.8 (estimated) 34.94
Standalone Asio / 1 context / 4 threads 233,780 [230,894, 235,048] 456.60 1,095.21 1,325.62 1,549.92 380.0 (estimated) 35.58

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 48,596 [41,893, 49,622] 1,518.62 5,139.58 6,142.54 6,740.75 199.4 (estimated) 47.42
Boost.Asio / 4 contexts / 4 threads 81,774 [78,667, 82,297] 2,555.43 3,086.67 3,442.96 3,578.46 397.4 (estimated) 71.70
Boost.Asio / 1 context / 1 thread 24,233 [24,111, 24,497] 757.28 10,453.83 11,265.50 12,300.83 98.3 (estimated) 47.03
Boost.Asio / 1 context / 2 threads 47,868 [45,187, 48,153] 1,495.89 5,346.04 5,952.96 6,364.79 198.7 (estimated) 47.39
Boost.Asio / 1 context / 4 threads 74,060 [68,925, 74,330] 2,314.37 3,460.42 4,189.96 4,492.17 390.7 (estimated) 47.39
Standalone Asio / 2 contexts / 2 threads 44,825 [41,184, 48,053] 1,400.77 5,604.33 6,659.12 6,968.25 198.5 (estimated) 47.61
Standalone Asio / 4 contexts / 4 threads 79,503 [78,050, 82,746] 2,484.46 3,229.62 3,519.17 3,603.33 398.1 (estimated) 47.66
Standalone Asio / 1 context / 1 thread 24,356 [23,595, 25,355] 761.12 10,243.71 11,459.33 11,601.12 98.3 (estimated) 47.62
Standalone Asio / 1 context / 2 threads 43,203 [42,987, 47,018] 1,350.10 5,830.12 6,996.71 7,441.54 197.8 (estimated) 47.47
Standalone Asio / 1 context / 4 threads 72,029 [66,510, 72,980] 2,250.90 3,535.29 4,623.58 5,210.58 389.4 (estimated) 47.64

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