/benchmark-pulling-proxy

Primary LanguageRuby

This project aims to benchmark how a polling proxy server perform when forwarding reqeusts to another slow response server using IO multiplexing architecture.

Start Server

docker-compose up --build

It starts 2 servers:

  1. slow response server - with 20-second response time.
  2. polling proxy - forwrad reqeust to slow response server and respond instantly with the the format in {"state":"pending"} or {"state":"done"}.

Sequence Diagram

Usage

curl 'http://localhost:3000?task_id=123'

output for first time:

{"state":"pending"}

output after 20 seconds using the same task_id:

{"state":"done","result":"ok"}

Benchmark

TL;DR: One single thread can handle approximately 2.5k concurrent clients.

Run the following command to benchmark:

./benchmark

example output:

Running 10s test @ http://polling-proxy:3000
  8 threads and 10000 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   280.47ms  117.21ms   1.70s    92.09%
    Req/Sec   362.93    419.94     2.02k    83.68%
  25179 requests in 10.09s, 2.72MB read
  Socket errors: connect 0, read 0, write 0, timeout 29
Requests/sec:   2494.44
Transfer/sec:    275.55KB