performance imporve

[thinkerou]

service resource have many rich.

Get the following data using Helloworld program which have not handle db/cache:

• Async: QPS < 2.5w
• Sync: QPS < 6k
• BUT use gatling test http, QPS > 5w

================================================================================
---- Global Information --------------------------------------------------------
> request count                                    1000000 (OK=1000000 KO=0     )
> min response time                                    100 (OK=100    KO=-     )
> max response time                                  39059 (OK=39059  KO=-     )
> mean response time                                 21825 (OK=21825  KO=-     )
> std deviation                                       6123 (OK=6123   KO=-     )
> response time 50th percentile                      22521 (OK=22522  KO=-     )
> response time 75th percentile                      24805 (OK=24858  KO=-     )
> response time 95th percentile                      28557 (OK=28556  KO=-     )
> response time 99th percentile                      32369 (OK=32368  KO=-     )
> mean requests/sec                                22222.222 (OK=22222.222 KO=-     )
---- Response Time Distribution ------------------------------------------------
> t < 800 ms                                         22199 (  2%)
> 800 ms < t < 1200 ms                               11107 (  1%)
> t > 1200 ms                                       966694 ( 97%)
> failed                                                 0 (  0%)
================================================================================

================================================================================
---- Global Information --------------------------------------------------------
> request count                                     200000 (OK=200000 KO=0     )
> min response time                                      0 (OK=0      KO=-     )
> max response time                                    453 (OK=453    KO=-     )
> mean response time                                     0 (OK=0      KO=-     )
> std deviation                                          1 (OK=1      KO=-     )
> response time 50th percentile                          0 (OK=0      KO=-     )
> response time 75th percentile                          0 (OK=0      KO=-     )
> response time 95th percentile                          1 (OK=1      KO=-     )
> response time 99th percentile                          1 (OK=1      KO=-     )
> mean requests/sec                                5882.353 (OK=5882.353 KO=-     )
---- Response Time Distribution ------------------------------------------------
> t < 800 ms                                        200000 (100%)
> 800 ms < t < 1200 ms                                   0 (  0%)
> t > 1200 ms                                            0 (  0%)
> failed                                                 0 (  0%)
================================================================================

[macchiatow]

Some performance degrades may be because of grpc server is on the same machine.

[thinkerou]

No, my test have two machine, one server, one client.
And two machine set export SBT_OPTS="-Xmx96G -XX:+UseConcMarkSweepGC -XX:+CMSClassUnloadingEnabled -Xss128M -Duser.timezone=GMT".

[macchiatow]

unfortunately I could solve to make it as fast as http

[thinkerou]

maybe it can help us https://github.com/gatling/gatling/blob/master/gatling-core/src/main/resources/gatling-akka-defaults.conf and https://github.com/akka/akka/blob/563c7fbcf0fcb374d414403900d620c8fac54d52/akka-bench-jmh/src/main/scala/akka/actor/AffinityPoolComparativeBenchmark.scala#L56 parallelism-factor, I will spend time see next week.

Because I can start two/three gatling process.