In order to confirm that there is not some nasty artefact inside of chronicle queue causing the nasty high percentile latencies we sometimes see, this is a test of chronicle queue against the simplest possible comparison I could find - Agrona’s ringbuffer.
Chronicle queue does a lot more than Agrona ringbuffer e.g. indexing, rolling, unbounded, multi-thread publish & consume etc.
There is a JLBH test for queue and another for Agrona and each test starts up a consumer thread to read messages, and generates them in the JLBH thread. The Agrona test uses a memory-mapped file to make it as similar as possible to CQ. We measure the time it takes to get from producer thread to consumer thread.
Agrona vs Queue
-
very similar at 99.99/worst
-
Agrona consistently better at lower percentiles which of course is to be expected as it is doing less
The fact that they are pretty much the same at worst says to me that we don’t have any lurking issues in chronicle queue.
TODO: test with 2 processes
-------------------------------- SUMMARY (end to end)------------------------------------------------------------
Percentile run1 run2 run3 run4 run5 run6 run7 % Variation
50: 0.19 0.14 0.14 0.14 0.14 0.15 0.16 7.08
90: 0.23 0.18 0.17 0.18 0.17 0.18 0.17 3.01
99: 0.28 0.20 0.20 0.20 0.20 0.20 0.20 2.65
99.9: 4.99 0.47 0.22 0.26 0.24 4.99 0.24 93.53
99.99: 16.13 7.81 8.45 8.06 8.45 9.47 8.45 12.44
worst: 96.26 33.79 10.50 10.50 16.13 10.50 10.50 59.67-------------------------------- SUMMARY (end to end)------------------------------------------------------------
Percentile run1 run2 run3 run4 run5 run6 run7 % Variation
50: 0.31 0.28 0.28 0.28 0.28 0.28 0.21 17.62
90: 0.34 0.33 0.31 0.33 0.31 0.31 0.26 13.91
99: 1.57 1.44 1.50 0.39 1.50 1.57 3.14 82.35
99.9: 6.53 5.50 4.99 4.03 8.45 7.81 9.47 47.35
99.99: 27.14 9.98 8.96 9.47 9.47 9.98 10.50 10.26
worst: 135.17 37.89 17.92 13.06 10.50 11.01 10.50 63.50
--------------------------------------------------------------------------------------------------------------------------------------------------- SUMMARY (end to end)------------------------------------------------------------
Percentile run1 run2 run3 run4 run5 run6 run7 % Variation
50: 0.20 0.16 0.15 0.16 0.15 0.16 0.16 6.72
90: 0.24 0.20 0.19 0.19 0.19 0.19 0.20 2.76
99: 0.28 0.24 0.21 0.21 0.21 0.22 4.99 93.76
99.9: 1.25 0.30 0.24 0.24 3.26 7.30 9.47 96.31
99.99: 12.54 8.45 8.45 8.06 8.06 9.98 10.50 16.74
worst: 129.02 92.16 10.50 10.50 46.08 11.01 12.03 83.84-------------------------------- SUMMARY (end to end)------------------------------------------------------------
Percentile run1 run2 run3 run4 run5 run6 run7 % Variation
50: 0.38 0.34 0.36 0.38 0.36 0.36 0.36 5.84
90: 0.42 0.39 0.41 0.41 0.39 0.41 0.41 2.65
99: 0.47 0.44 0.46 0.46 0.46 0.47 4.03 84.48
99.9: 0.91 0.82 0.53 0.53 2.37 7.81 9.47 91.87
99.99: 10.50 10.50 9.47 9.47 8.96 10.50 10.50 10.26
worst: 135.17 135.17 135.17 135.17 135.17 135.17 96.26 21.23These tests compare purely performance for a larger, longer-running workload. Each run is 5,000,000 iterations of a 128 byte message @ 50K/sec = 620MB of data.
-------------------------------- SUMMARY (end to end)---------------------------------
Percentile run1 run2 run3 run4 run5 % Variation
50: 0.34 0.33 0.33 0.33 0.33 0.00
90: 0.41 0.39 0.39 0.39 0.39 0.17
99: 3.04 0.47 1.46 2.03 1.73 69.17
99.9: 5.13 3.47 3.35 3.38 3.37 2.41
99.99: 2806.78 4.53 3.61 3.71 3.61 14.45
99.999: 11554.82 452.22 593.15 734.46 7.96 98.38
worst: 12513.28 1422.85 1555.97 1690.11 362.62 70.93-------------------------------- SUMMARY (end to end)---------------------------------
Percentile run1 run2 run3 run4 run5 % Variation
50: 0.59 0.59 0.59 0.59 0.59 0.11
90: 0.67 0.66 0.66 0.66 0.66 0.20
99: 3.39 3.42 3.42 3.39 3.39 0.51
99.9: 300.16 43.73 7.06 4.31 4.72 85.91
99.99: 17473.54 6109.18 4945.92 4191.23 4581.38 23.38
99.999: 25649.15 13643.78 6084.61 5883.90 5908.48 46.79
worst: 26386.43 14397.44 6498.30 6195.20 6428.67 46.88-------------------------------- SUMMARY (end to end)----------------------------------
Percentile run1 run2 run3 run4 run5 % Variation
50: 0.61 0.62 0.63 0.63 0.63 0.53
90: 0.66 0.66 0.67 0.67 0.67 0.20
99: 1.81 0.73 0.72 0.73 0.71 1.84
99.9: 7.43 1.86 1.85 1.95 0.78 49.93
99.99: 17637.38 5421.06 2.02 4.08 1.00 99.97
99.999: 26501.12 9826.30 383.87 267.39 43.79 99.33
worst: 27451.39 10465.28 1142.27 1137.15 561.41 92.16
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent