Welcome to Centmin Mod Community
Become a Member

WPPerformanceTester Results

Discussion in 'Blogs & CMS usage' started by tininho, Jun 24, 2019.

  1. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    I am simply wondering how the other stack on similar Upcloud server is optimized to achieve that performance on PHP bench. It seems Centminmod scores constantly better on WP Bench, but PHP bench does not improve or is worse than Webinoly out of the box, for example.


    Upcloud 1CPU 2GB HEL1 /w Centminmod VS
    benchmark stacks/providers



    WPPerformanceTester Results

    WPPerformanceTester performs two benchmarks. One is a WordPress (WP Bench) and the other a PHP benchmark. WP Bench measures how many WP queries per second and higher tends to be better (varies considerably by architecture). PHP Bench performs a lot of computational and some database operations which are measured in seconds to complete. Lower PHP Bench is better.


    (PHP Bench & WP Bench)

    Servebolt (not Upcloud, fastest PHP Bench): 3.809
    & 1680.67
    Local fastest* Wordpress hosting (Upcloud similar server, custom stack): 6.562 & 343

    Webinoly (Upcloud 1CPU2GB):
    8.218 & 881.83
    Centminmod (Upcloud 1CPU2GB): 8.304 & 1795.33
    maybe more later



    * Claiming so.
    Centmin: Latest with Insight Guide - How to boost Centmin Mod LEMP stack performance and the latest custom_config.inc recommendations from this forum
    Servebolt: $51-100/Month WordPress Hosting Performance Benchmarks (2019) | Review Signal Blog
     
    Last edited: Jun 25, 2019
    • Like Like x 1
    • Informative Informative x 1
  2. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    Since the average score for the PHP bench seems to be between 8-10 sec, it is interesting to find out how they (especially Servebolt) pushed for a lower timing (=faster=better) score.

    Either they went the "Dieselgate" way or they are relatively good at optimizing things.

    If I got this right, more than 90% of this PHP bench is a battle on this test:

    Mysql Query Benchmark
    Time it takes for 1,000,000 ENCODE()s with a random seed​
     
    Last edited: Jun 25, 2019
  3. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    style="display:inline-block;min-width:400px;max-width:970px;width:95%;height:90px" data-ad-client="ca-pub-6669518204467592" data-ad-slot="4024536743" data-ad-format="auto">
  4. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    I am comparing WPPerformanceTester results between different stacks (Centminmod, Webinoly etc.) on Upcloud 1CPU 2GB against Servebolt & local "fastest" Wordpress hosting on Upcloud.

    So the server hardware used here is Upcloud 1CPU 2GB

    (except Servebolt, which server specs are unknown and priced 100EUR/m, but is the FASTEST. As it does not seem you can get faster (on WPPerformanceTester) by throwing more money in (according to reviewsignal.com), it seems reasonable to benchmark against, no?)

    Centminmod is relatively the fastest stack, but I can't figure out how they (Servebolt and local provider) managed to get that result on PHP Bench. This is the point.

    Also it seems PHP Bench is what gives the least marginal for Centminmod against other stacks, or even looses.
     
    Last edited: Jun 25, 2019
  5. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    Yeah without knowing what Serverbolt uses hardware wise it would be hard though it's pretty easy to guess as php performance is closely tied to cpu clock speed, higher clocks = faster PHP performance usually when we're talking >50% faster speeds. Anything within 1-3% deviation is usually minor variance within norms for that PHP bench script. Also dedicated servers or VPS with higher cpu dedicated percentage allocations will be faster than VPS servers with less cpu % allocations.

    Though you'd rather have higher WP Bench scores as that speaks to underlying real performance without any wordpress related caching methods to help. Whereas PHP Bench related performance can be bypassed somewhat with Cloudflare accelerator proxies and/or Wordpress full page caching methods which allow WP visitor loads to not even touch PHP so it all becomes about Nginx static full page caching performance usually i.e. better indicator of performance in terms of HTTP/2 HTTPS load/concurrency
     
  6. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    Yeah without knowing what Serverbolt uses hardware wise it would be hard though it's pretty easy to guess as php performance is closely tied to cpu clock speed, higher clocks = faster PHP performance usually when we're talking >50% faster speeds.
    [...]
    PHP Bench related performance can be bypassed somewhat with Cloudflare accelerator proxies and/or Wordpress full page caching methods

    I can't seem to reach this PHP Bench score with the caching methods I am familiar with. Actually nothing I did had any effect on the score. As this result correlates with my skills on the matter, I am not surprised.

    It does not, however, correlate with the time I have used for this, so I would greatly appreciate if you or anyone else could nod to the right direction.
     
  7. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    No wp cache or server cache will allow you to reduce PHPbench performance from say 8-11 seconds to 3.5-4 seconds - 95% sure it's server hardware/% dedicated resource allocation mix related.

    You can try your wpperformancetester tests on Centmin Mod 123.09beta01 on Vultr High Frequency VPS clocked at 3.8Ghz instead and see how it fairs.
     
  8. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    right from Serverbolt's own mouth so to speak Test your WordPress Webhost's Performance with WP Performance Tester
     
  9. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    2018 Servebolts results:
    Servebolt.com Standard 8.445 & 1412.4
    Servebolt Mini Dedicated 6.741 & 2092.0

    2019 Servebolts results
    3.809
    & 1680.67

    So it seems they have found a way to improve both results. They are claiming (on their website) they have custom build php libraries etc. to achieve this. On 2019 test which they won, they have implied that the 2nd best boosted their CPU for this test. This gives me a reason to believe this is not a "Dieselgate" (they don't cheat).

    I have personally tested Hetzner's EX62-NVME (Intel® Core™ i9-9900K Octa-Core ), results
    5 & 3600 (around, this server is not currently active, have to recheck)

    FYI, the 3x more expensive AX160-NVME with AMD EPYC 7401P 24 Cores "Naples" (Zen)
    scored way less (~20-25%) a month ago, but I can't remember the exact numbers (have to recheck).

    Both with a similar Centminmod setup as mentioned above.

    Now, since the above mentioned fastest* local Wordpress hosting (Upcloud, similar server, custom stack) scores
    6.562 & 343 (range is 180-350 for WP bench, PHP Bench is pretty stable) -

    I am led to believe that there is something I could do to improve PHP Bench on Upcloud 1CPU2GB.
    Even if it's a tradeoff between the two, somehow.
     
    Last edited: Jun 25, 2019
  10. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
  11. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    On Upcloud 1CPU 1GB FIN:

    Your settings replicated from above post and add/modify to custom_config.inc prior to rebuilding with option 5:

    Code:
    PHPPGO_INDEXPATH='home/nginx/domains/mydomain/public/wp-content/plugins/wpperformancetester/benchmark.php'
    PHPPGO_INDEXPATH_TWO='/home/nginx/domains/mydomain/public/index.php'
    
    PHP_PGO='y'
    PHP_PGO_ALWAYS='y'
    https://raw.githubusercontent.com/tinof/centmininit/master/custom_config.inc

    Result: No observable change in the benchmark.

    PHP Upgrade log (benchmark.php pgo):

    Code:
    ./sapi/cgi/php-cgi -T100 /home/nginx/domains/mydomain/public/wp-content/plugins/wpperformancetester/benchmark.php
    Content-type: text/html; charset=UTF-8
    
    Content-type: text/html; charset=UTF-8 (this same lane for 80 times)
    
    Elapsed time: 0.004347 sec
    
    real    0m0.067s
    user    0m0.046s
    sys    0m0.020s

    This is the part from benchmark.php, in which Centminmod (8.3 sec) loses to Webinoly (<8sec sec, have to reconfirm) and
    local Upcloud hosting (6.5 sec):


    Full benchmark.php file: benchmark.php

    So currently I can't replicate this result (<8sec in PHPBench on Upcloud 1CPU1GB).
     
    Last edited: Jul 12, 2019
  12. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    1st path has invalid missing forward slash at front not that it might make a difference

    Which version of php? 7.3.7 is the latest

    Also for respective lemp stacks what is output for command
    Code (Text):
    php-config
     
  13. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    Fair point. This was a mistake in the post, I checked the log file and the path is correct (/home/nginx/domains/mydomain/public/wp-content/plugins/wpperformancetester/benchmark.php)

    As for the PHP related question:

    --version [7.3.7]
    php-config php-config

    EDIT: I was wrong about Webinoly vs Centminmod. Upcloud changes processors between Intel(R) Xeon(R) CPU E5-2687W (faster) v4 @ 3.00GHz and Intel(R) Xeon(R) CPU E5-2687W v3 @ 3.10GHz
     
    Last edited: Jul 12, 2019
  14. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    and for webinoly and local fast upcloud's php-config output ? allows comparing their php-fpm configurations in terms of what's configured and what extensions and modules are loaded
     
  15. fabianski

    fabianski Member

    94
    9
    8
    Feb 20, 2019
    Brazil
    Ratings:
    +28
    Local Time:
    5:08 PM
    • Informative Informative x 1
  16. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    Yeah so you can't really compare php-fpm with mod_php due to php-fpm's TCP network overhead. Though i wouldn't focus too much on php bench scripts synthetics when you should be testing php requests via the web server over HTTPS i.e. PHP - PHP 7.x Benchmarks Centmin Mod vs Easyengine vs Webinoly vs VestaCP vs OneInStack as that factors in the whole LEMP/LAMP stack from php processing and passing data onto Nginx/webserver and web server processing that and passing it onto visitors.
     
  17. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    @tininho just setup a new Wordpress site on $5/month 1GB Upcloud.com KVM VPS with Centmin Mod 123.09beta01 with PHP 7.3.7 default via centmin.sh menu option 22 and did WPPerformanceTester WP Plugin run and looks good here with Intel Xeon Gold 6136 in Upcloud Chicago datacenter. Testing centmin.sh menu option 22's upcoming PHP-FPM fastcgi_cache full page caching option :)

    upcloud-5m-wpperformancetester-130719-01.png
     
    • Winner Winner x 1
  18. tininho

    tininho Premium Member Premium Member

    18
    1
    3
    May 22, 2019
    Ratings:
    +7
    Local Time:
    11:08 PM
    Nginx Ver: 1.17
    MariaDB Ver: 10.4
    @eva2000

    Very interesting, thank you. It seems Chicago has faster CPUs than EU-servers (can't test myself, I get error from Upcloud saying no free IP's available when I choose CHI1).

    So here are the current results (Centminmod setup from above):

    Hetzner EX 62-NME (Intel® Core™ i9-9900K: @ 3.60GHz)
    4.7 / 3600

    Vultr 1CPU1GB Amsterdam (Intel @ 3.8GHz)
    6.1 / 2400

    Upcloud 1CPU1GB CHI1 (Intel Xeon Gold 6136 @ 3.00Ghz)
    6.5 / 2100
    This from @eva2000.


    Upcloud 1CPU1GB HEL1/FRA/AMS (Intel(R) Xeon(R) CPU E5-2687W [email protected] 3.00GHz)
    7.5 / 1850

    Upcloud 1CPU1GB HEL2 (Intel(R) Xeon(R) CPU E5-2687W v3 @ 3.10GHz)
    8.3 / 1850

    The numbers are highest/lowest score I get, not average.

    It seems there is no way around the fact that this test is CPU bound (like you said). I assume the forementioned local hosting has simply these faster CPU models on their stack (even though they are not publically available in EU?).

    PHP/Mysql tuning, PHP PGO nor FAST-CGI caching had no effect on this what so ever.

    Well, next is MySQL 8 testing :)
     
    Last edited: Jul 15, 2019 at 9:54 PM
    • Informative Informative x 2
  19. fabianski

    fabianski Member

    94
    9
    8
    Feb 20, 2019
    Brazil
    Ratings:
    +28
    Local Time:
    5:08 PM
    I think this is for everyone, i'm getting the same error
     
  20. eva2000

    eva2000 Administrator Staff Member

    40,634
    9,023
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +13,891
    Local Time:
    6:08 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    same cpus just VPS are shared resources still so depends on individual VPS host nodes overall load from all VPS users on the same server. Only dedicated servers with dedicated resources would have a more accurate result.

    i guess i was lucky getting my Chicago VPS then