Github user HeartSaVioR commented on the issue:
https://github.com/apache/storm/pull/2261
@revans2
Thanks for doing the performance test. I just ran some tests (rate 100000,
1 worker, max spout pending 1000) with TVL and got below result. Picked only
240 secs, but I also have raw numbers as well.
> patch
uptime: 240 acked: 3,014,700 acked/sec: 100,490.00 failed: 0 99%:
14,041,087 99.9%: 17,924,095 min: 3,284,992 max: 30,244,863
mean: 7,576,927.04 stddev: 1,771,441.20 user: 120,360 sys: 5,830
gc: 1,454 mem: 135.74
uptime: 241 acked: 3,013,460 acked/sec: 100,448.67 failed: 0 99%:
13,623,295 99.9%: 16,859,135 min: 3,282,944 max: 24,559,615
mean: 7,498,565.53 stddev: 1,626,489.06 user: 119,890 sys: 5,550
gc: 1,320 mem: 73.44
> master
uptime: 240 acked: 3,013,700 acked/sec: 100,456.67 failed: 0 99%:
13,533,183 99.9%: 17,563,647 min: 3,311,616 max: 25,821,183
mean: 7,470,969.09 stddev: 1,618,231.68 user: 118,480 sys: 5,720
gc: 1,236 mem: 97.66
uptime: 241 acked: 3,015,020 acked/sec: 100,500.67 failed: 0 99%:
13,516,799 99.9%: 16,654,335 min: 3,284,992 max: 22,331,391
mean: 7,501,903.32 stddev: 1,650,650.14 user: 119,170 sys: 5,320
gc: 1,364 mem: 84.73
> patch, acker off
uptime: 240 acked: 3,013,040 acked/sec: 100,434.67 failed: 0 99%:
4,464,639 99.9%: 6,959,103 min: 283 max: 11,902,975
mean: 634,505.19 stddev: 696,511.91 user: 76,820 sys: 6,470
gc: 868 mem: 165.32
uptime: 240 acked: 3,012,520 acked/sec: 100,417.33 failed: 0 99%:
4,255,743 99.9%: 6,008,831 min: 323 max: 9,650,175
mean: 627,677.58 stddev: 651,341.18 user: 76,300 sys: 6,860
gc: 885 mem: 111.09
> master, acker off
uptime: 241 acked: 3,013,420 acked/sec: 100,447.33 failed: 0 99%:
2,060,287 99.9%: 4,468,735 min: 356 max: 7,733,247
mean: 565,479.02 stddev: 417,299.37 user: 73,570 sys: 7,070
gc: 399 mem: 237.66
uptime: 241 acked: 3,012,800 acked/sec: 100,426.67 failed: 0 99%:
3,928,063 99.9%: 7,241,727 min: 450 max: 17,530,879
mean: 617,198.01 stddev: 658,032.31 user: 75,270 sys: 7,290
gc: 731 mem: 241.72
I've shown numbers from grouper performance test, and also
ConstSpoutNullBoltTopo which clearly shows newer is (much) faster, so actually
the result is confusing to me. Maybe better to put more efforts to standardize
approach of performance tests. I'll try to find time to have a look at loadgen.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
