OlegDB

Benchmarking with Wrk

2014-04-12 by Quinlan PfifferComments

Today at Proleg Labs™ we're going to discuss the awesome bechmarking tool wrk. wrk is a badass little HTTP stress testing utility written by a mysterious open-source do-gooders. I'm gonna run through how we torture OlegDB with it to make sure everything is working in an acceptable fashion.

Building and installing wrk is outside the scope of this article, but it's not hard. You just need to do your standard make && make install dance and you'll do fine. Once you have your wrk executable handy, we can get started.

On my particular machine, I like to test with these settings:

$ ./wrk -t8 -c400 -d30s http://localhost:8080/test

This requires a little explanation.

• -t8 is the number of threads to use. I've got eight cores, lets use eight threads.
• -c400 is the number of connections to have open. 400 seems like a nice, round number.
• -d30s is the amount of time to stress test. Thirty seconds should be enough for anyone.

and the last parameter is the URL we'll be hitting with the wrk jackhammer. Since we need to actually test Oleg and not some 404 page we'll need to put something into the database for wrk to fetch. Results will vary based on how much data you put in, but lets just test with this quote from Thomas Hobbes' Leviathan:

Man is distinguished, not only by his reason, but by this singular passion from
other animals, which is a lust of the mind, that by a perseverance of delight
in the continued and indefatigable generation of knowledge, exceeds the short
vehemence of any carnal pleasure.

Say, thats neat Thomas Hobbes! Let's throw that into OlegDB, which I'll assume you have running on localhost:8080:

$ curl -X POST -d '
> Man is distinguished, not only by his reason, but by this singular passion from
> other animals, which is a lust of the mind, that by a perseverance of delight
> in the continued and indefatigable generation of knowledge, exceeds the short
> vehemence of any carnal pleasure.' http://localhost:8080/test
無駄

Just to do a quick test that it actually inserted (Hey, who knows, right?) let's curl it back out:

$ curl localhost:8080/test
Man is distinguished, not only by his reason, but by this singular passion from
other animals, which is a lust of the mind, that by a perseverance of delight
in the continued and indefatigable generation of knowledge, exceeds the short
vehemence of any carnal pleasure.

Excellent, now lets fire up wrk, point it at that URL and see what happens:

$ ./wrk -t8 -c400 -d30s http://localhost:8080/test
Running 30s test @ http://localhost:8080/test
  8 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    18.05ms   93.64ms 866.52ms   96.72%
    Req/Sec     4.71k     2.15k   16.44k    70.58%
  1070154 requests in 30.00s, 420.48MB read
  Socket errors: connect 0, read 0, write 0, timeout 27
Requests/sec:  35675.16
Transfer/sec:     14.02MB

Hey, 35,600 Requests/second. Not bad. Not bad at all. For fun, let's see how Kyoto Tycoon performs with the same data and conditions:

$ ./wrk -t8 -c400 -d30s http://localhost:1978/test
Running 30s test @ http://localhost:1978/test
  8 threads and 400 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     3.05ms    2.47ms 340.86ms   99.65%
    Req/Sec    17.59k     2.75k   37.67k    77.99%
  3991195 requests in 30.00s, 1.39GB read
Requests/sec: 133041.46
Transfer/sec:     47.58MB

...well damn. I guess theres always room to improve.