Join the community today
Register Now

loader.io LOW? score

Discussion in 'Web Development & Web Performance' started by GASTAN, Dec 29, 2018.

  1. GASTAN

    GASTAN Member

    72
    11
    8
    Jun 28, 2017
    Ratings:
    +16
    Local Time:
    1:28 AM
    Hi

    I have
    CentOS Linux release 7.6.1810 (Core)
    x86_64 x86_64 x86_64 GNU/Linux
    123.09beta01
    nginx/1.15.8 (271218-074410)
    PHP 7.3.0 (cli) (built: Dec 20 2018 22:58:49) ( NTS )
    Zend Engine v3.3.0-dev, with Zend OPcache v7.3.0
    mysqld Ver 10.1.37-MariaDB for Linux on x86_64 (MariaDB Server)

    OVZ with 4GB RAM

    After testing my WP site with loader.io I got the feeling, it's quite slow compared to wordpress7.centminmod.com, which can handle 10 000 visitors, while my site seems to only take around 6k (without timing out)

    Any ideas on what could be wrong, or if I actually interpreted results right?

    thx
    --G
     

    Attached Files:

  2. GASTAN

    GASTAN Member

    72
    11
    8
    Jun 28, 2017
    Ratings:
    +16
    Local Time:
    1:28 AM
    uloaded wrong file before blog.test.responses.jpg
     
  3. eva2000

    eva2000 Administrator Staff Member

    42,749
    9,670
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +14,925
    Local Time:
    10:28 AM
    Nginx 1.17.x
    MariaDB 5.5/10.x
    loader.io randomly tests from different geographic locations from their testers so distance between your server and loader.io tester server can determine base response round trip time which in turn determins how many requests you can process over a fixed period of time. So if average ping response time is 500ms or 0.5s, over 10 seconds you can handle roughly 10/0.5 = 20 requests. If average ping response time is 50ms or 0.05s, over 10 seconds you can handle roughly 10/0.05 = 200 requests.

    So make sure loader.io test server geographic location distance to target test server is known similar to other compared loader.io tests. Otherwise, you're comparing apples to oranges as the distance maybe greater or less and thus average ping response time would vary and affect overall results too.

    Also OpenVZ VPS can not do TCP Kernel level tuning to optimise server concurrency handling as you are subject to using OpenVZ host node's TCP Kernel settings. KVM/XEN/VMWare and dedicated servers use their own Kernel and can do TCP Kernel level tuning.

    As per Centmin Mod install notes at Centmin Mod LEMP Stack Install Nginx on CentOS, as such Centmin Mod 123.09beta01 and higher auto detects if you use non-OpenVZ VPS and auto tune TCP Kernel settings for better user concurrency scaling and skips such auto tuning for OpenVZ VPS detected systems.

    If you want to handle more concurrent users, do not use OpenVZ based VPS systems.
     
  4. GASTAN

    GASTAN Member

    72
    11
    8
    Jun 28, 2017
    Ratings:
    +16
    Local Time:
    1:28 AM
    Oh, I did not realize you server is KVM. Prolly makes sense if you want performance. Well, maybe in future we get KVM, but not there yet. It seems it could be quite a big diferrence between OVZ and KVM then?
    I actually just wanted to benchmark site somehow before moving to new server, to see if it's faster. Would you recommend some other way?

    I just tried site in Q and it does not look good at all:
    upload_2018-12-29_15-34-30.png

    It breaks the test

    upload_2018-12-29_15-35-20.png

    Could it be this machine only has 1GB RAM?
    anyway, where is location of test from option?
     
  5. eva2000

    eva2000 Administrator Staff Member

    42,749
    9,670
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +14,925
    Local Time:
    10:28 AM
    Nginx 1.17.x
    MariaDB 5.5/10.x
    you'd have to test at lower concurrency levels within what your OpenVZ is capable of without timeouts but still keep in mind loader.io test server geographic location is same distance between old vs new VPS tests