Akka-nbench is a benchmarking tool created to conduct accurate and repeatable performance tests and stress tests, and produce performance graphs.
AKka-nbench focuses on two metrics of performance:
- Throughput: number of operations performed in a timeframe, captured in aggregate across all operation types
- Latency: time to complete single operations, captured in quantiles per-operation
See TODO
You first need to have JDK and Typesafe Activator.
You should install R for Graph Generation.
$ sudo apt-get install r-base
$ R
> install.packages(c("getopt", "ggplot2", "reshape", "plyr", "proto", "digest"))
> q()
$ git clone https://github.com/kotakanbe/akka-nbench.git
$ cd akka-nbench
$ activator -mem 512 "run-main bench.Bench sleep"
$ activator -mem 512 "run-main bench.Summarizer"
$ make results
$ open tests/current/summary.png
https://github.com/kotakanbe/akka-nbench/blob/master/src/main/resources/application.conf
sleep { # Scenario_name
concurrent = 100 # Sec
duration = 10 # Sec
driver = "bench.drivers.SleepDriver" # FQCN of Driver Class
operations = {
"sleep1" = 6 # operation_name = ratio
"sleep2" = 4 # operation_name = ratio
}
}
If you defined as above, it means
- by 100 Threads(Akka actor)
- for 10 Seconds
- The Driver Class is bench.drivers.SleepDriver
- sleep1 ... 60 Threads ( 6 / (6 + 4) * 100 )
- sleep2 ... 40 Threads ( 4 / (6 + 4) * 100 )
$ cd akka-nbench/src/main/scala/drivers
$ cp SleepDriver.scala MyDriver.scala
$ vim
akka-nbench/src/main/scala/drivers/MyDriver.scala
- ClassName
class MyDriver(...)
- Method Contents
def sleep3(): (Boolean, Long, Long) = {
val start = System.currentTimeMillis
log.info("sleep3")
Thread.sleep(1000)
val endAt = System.currentTimeMillis
val elapsedMillis= endAt - start
(true, endAt, elapsedMillis)
}
- getOperataion method
override val getOperation = () => {
operation match {
case "sleep3" => sleep3 _
}
}
akka-nbench/src/main/resources/application.conf
mydriver {
concurrent = 10
duration = 10
driver = "bench.drivers.MyDriver"
operations = {
"sleep3" = 1
}
}
Benchmarking
$ cd akka-nbench
$ activator -mem 512 "run-main bench.Bench mydriver"
$ activator -mem 512 "run-main bench.Summarizer"
$ make results
$ open tests/current/summary.png
Benchmaking on Multi Clients and Collect raw.csv. Then cat and sort...
$ cat raw1.csv raw2.csv raw3.csv | sort > raw.csv
$ activator -mem 512 "run-main bench.Summarizer"
$ make results
$ open tests/current/summary.png
Unix/Linux