case1.md

Network IO, Case 1

• build a delay service which will return {time} ms with /delay/ms/{time} endpoint
• the target server will invoke delay endpoint and return

Environment

• AWS t2.xlarge (4core 16g) -> Docker, Spring Web (Delay Service)
• AWS t2.xlarge (4core 16g) -> Docker, Spring Web, Spring Web Async
• AWS t2.xlarge (4core 16g) -> Docker, Spring Reactive Web, Spring Reactive Web Coroutine
• AWS t2.xlarge (4core 16g) -> Docker, Vert.x Verticle
• AWS t2.xlarge (4core 16g) -> Docker, Vert.x Coroutine Verticle
• AWS t2.xlarge (4core 16g) -> Docker, Ktor
• AWS t2.2xlarge (8core 32g) -> openjdk-11-jdk, Apache-JMeter-5.4.1

Context

• target server receive request and invokes /delay/ms/500, /delay/ms/800, /delay/ms/1000 endpoints concurrently then return
• jmeter use constant throughput timer to keep 20 RPS
• for demo-delay-service

server.tomcat.threads.max=800

• for demo-spring-mvc

server.tomcat.threads.max=800

• for demo-spring-flux

-Dreactor.netty.ioWorkerCount=1000 -Dreactor.netty.pool.maxConnections=8192

• for demo-vertx

vertx.deployVerticle(
    "com.example.demo.vertx.VertxVerticle",
    DeploymentOptions().setInstances(VertxOptions.DEFAULT_EVENT_LOOP_POOL_SIZE)
)

vertx.deployVerticle(
  "com.example.demo.vertx.ServiceVerticle", 
  DeploymentOptions()
    .setInstances(VertxOptions.DEFAULT_EVENT_LOOP_POOL_SIZE)
    .setWorker(true)
    .setWorkerPoolSize(1000)
)

• for demo-ktor

ktor {
    deployment {
        callGroupSize = 1000
        connectionGroupSize = 1000
        workerGroupSize = 1000
    }
}

Procedure

• Warm twice and hit one once

Spring Web

Spring Web Async

Spring Reactive Web

Spring Reactive Web Coroutine

Vert.x

Vert.x Coroutine

Ktor