[
https://issues.apache.org/jira/browse/DISPATCH-2162?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17371442#comment-17371442
]
Ken Giusti commented on DISPATCH-2162:
--------------------------------------
Improved performance a bit via [this
commit|[https://github.com/apache/qpid-dispatch/commit/962da1e/].]
{{$
~/work/dispatch/dispatch-tester/benchmarks/test-configurations/http1-perf/wrk/wrk
-c 16 -t 4 -d20s http://127.0.0.1:8000/t10K.html }}
{{Running 20s test @ http://127.0.0.1:8000/t10K.html }}
{{ 4 threads and 16 connections }}
{{ Thread Stats Avg Stdev Max +/- Stdev }}
{{ Latency 2.51ms 148.86us 7.25ms 91.62% }}
{{ Req/Sec 1.60k 30.20 1.77k 91.12% }}
{{ 127228 requests in 20.00s, 1.33GB read }}
{{Requests/sec: 6359.89 }}
{{Transfer/sec: 67.86MB}}
While there's always more room for improvement this will have to do for 1.17.0
as I'm refocusing the HTTP/1.x work to improving stability and fixing bugs.
> HTTP/1.x throughput is abysmal.
> -------------------------------
>
> Key: DISPATCH-2162
> URL: https://issues.apache.org/jira/browse/DISPATCH-2162
> Project: Qpid Dispatch
> Issue Type: Bug
> Components: Protocol Adaptors
> Affects Versions: 1.16.0
> Reporter: Ken Giusti
> Assignee: Ken Giusti
> Priority: Critical
> Labels: HTTP/1.x
> Fix For: 1.17.0
>
>
> The throughput of the HTTP/1.x adaptor is so bad it is basically unusable.
> Using the _wrk_ HTTP load generator against an nginx server running in a
> container and serving 10 KB response "pages":
>
> $ ./wrk/wrk -c 16 -t 4 -d20s http://127.0.0.1:8080/t10K.html
> Running 20s test @ http://127.0.0.1:8080/t10K.html
> 4 threads and 16 connections
> Thread Stats Avg Stdev Max +/- Stdev
> Latency 554.12us 292.27us 13.51ms 86.10%
> Req/Sec 7.21k 1.57k 19.57k 77.90%
> 575136 requests in 20.10s, 6.01GB read
> Requests/sec: 28614.93
> Transfer/sec: 306.00MB
>
> Same test running across a single qdrouterd instance:
>
> [kgiusti@localhost http1-perf (master)]$ ./wrk/wrk -c 16 -t 4 -d20s
> http://127.0.0.1:8000/t10K.html
> Running 20s test @ http://127.0.0.1:8000/t10K.html
> 4 threads and 16 connections
> Thread Stats Avg Stdev Max +/- Stdev
> Latency 14.51ms 13.93ms 244.87ms 98.68%
> Req/Sec 302.50 39.30 370.00 76.13%
> 23998 requests in 20.02s, 256.07MB read
> Requests/sec: 1198.81
> Transfer/sec: 12.79MB
>
> See
> [https://github.com/kgiusti/dispatch-tester/tree/master/benchmarks/test-configurations/http1-perf]
> for test setup details.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
