libuv_http_benchmark.md

Description

I decided to hook up the node.js C libuv async IO library that uses IOCP on Windows and epoll/kqueue/event ports/etc. on Unix systems to handle HTTP traffic to test if this is a suitable HTTP front end for Dazzle.

There is no V8 javascript engine so it's not running node.js, only the libuv async IO eventing and some C for handling the HTTP parsing.

Setup

• libuv is using default settings, zero tuning has been done.
• Client requests coming from a single remote client machine (Linux) over the LAN.
• Dazzle + libuv running on Windows 7 Core i7 930, 16GB of RAM.
• Dazzle returns "Hello world" in the HTTP response.

During the tests - CPU: 10% RAM: ~1.5MB

Results (Wrk HTTP Pipelining)

Wrk is a HTTP multi-threaded benchmark tool that supports HTTP pipelining.

wrk -d10 -t8 -c1024 --pipeline 64 http://192.168.0.201:8000/

Result: 291,346.23 requests per second with 0.0033% lost requests.

Running 10s test @ http://192.168.0.201:8000/
  8 threads and 1024 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency    73.66ms   76.62ms 516.61ms   91.31%
    Req/Sec    35.20k    12.70k   62.00k    57.53%
  2915204 requests in 10.01s, 283.58MB read
  Socket errors: connect 11, read 0, write 0, timeout 965

Requests/sec: 291,346.23
Transfer/sec:     28.34MB

Results (Wrk non-pipelined)

wrk -d10 -t8 -c32 http://192.168.0.201:8000/

Result: 65,186.19 requests per second.

Running 10s test @ http://192.168.0.201:8000/
  8 threads and 32 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency   829.92us    1.65ms  14.94ms   94.91%
    Req/Sec     8.04k   343.23     9.00k    90.19%
  651841 requests in 10.00s, 63.41MB read
  
Requests/sec:  65,186.19
Transfer/sec:      6.34MB

Results (Wrk with higher load)

If we crank the concurrent connections way up and increase threads Dazzle gains 10,000 more requests/second but now the load is affecting latency.

wrk -d10 -t16 -c512 http://192.168.0.201:8000/

Result: 75,045.61 requests per second.

Running 10s test @ http://192.168.0.201:8000/
  16 threads and 512 connections
  Thread Stats   Avg      Stdev     Max   +/- Stdev
    Latency     8.38ms    8.61ms  61.37ms   86.81%
    Req/Sec     4.16k   457.19     8.00k    79.24%
  749530 requests in 9.99s, 72.91MB read
  
Requests/sec:  75,045.61
Transfer/sec:      7.30MB

Results (Apache Bench - no keep-alive)

ab -n 100000 -c 32 http://192.168.0.201:8000/

Result: 12,085.06 requests per second.

Server Hostname:        192.168.0.201  
Server Port:            8000  

Document Path:          /  
Document Length:        13 bytes  

Concurrency Level:      64  
Time taken for tests:   8.275 seconds  
Complete requests:      100000  
Failed requests:        0  
Write errors:           0  
Total transferred:      7800000 bytes  
HTML transferred:       1300000 bytes  
Requests per second:    12,085.06 [#/sec] (mean)  
Time per request:       5.296 [ms] (mean)  
Time per request:       0.083 [ms] (mean, across all concurrent requests)  
Transfer rate:          920.54 [Kbytes/sec] received  

Connection Times (ms)  
              min  mean[+/-sd] median   max  
Connect:        0    2   0.9      2      16  
Processing:     1    3   0.8      3      12  
Waiting:        1    2   0.7      2      11  
Total:          3    5   1.4      6      26  

Percentage of the requests served within a certain time (ms)  
  50%      6  
  66%      6  
  75%      6  
  80%      6  
  90%      6  
  95%      7  
  98%      7  
  99%      8  
 100%     26 (longest request)  

Results (Apache Bench - keep-alive)

ab -n 100000 -c 32 -k http://192.168.0.201:8000/

Result: 33,314.89 requests per second.

Server Hostname:        192.168.0.201
Server Port:            8000

Document Path:          /
Document Length:        13 bytes

Concurrency Level:      32
Time taken for tests:   3.002 seconds
Complete requests:      100000
Failed requests:        0
Write errors:           0
Keep-Alive requests:    100000
Total transferred:      10200000 bytes
HTML transferred:       1300000 bytes
Requests per second:    33,314.89 [#/sec] (mean)
Time per request:       0.961 [ms] (mean)
Time per request:       0.030 [ms] (mean, across all concurrent requests)
Transfer rate:          3318.48 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       2
Processing:     0    1   0.3      1       5
Waiting:        0    1   0.3      1       5
Total:          0    1   0.3      1       5

Percentage of the requests served within a certain time (ms)
  50%      1
  66%      1
  75%      1
  80%      1
  90%      1
  95%      1
  98%      2
  99%      2
 100%      5 (longest request)

awesome!

Great!