Want more timely Centmin Mod News Updates?
Become a Member

Vultr Bare Metal vs. High Frequency (Vultr) for 3 small forums & a couple WordPress sites

Discussion in 'Virtual Private Server (VPS) hosting' started by frm, Sep 2, 2020.

  1. frm

    frm Member

    33
    4
    8
    Dec 17, 2018
    Ratings:
    +5
    Local Time:
    9:43 PM
    1.15.7
    10.3.11
    I'm currently running 1 XenForo forum with a HF 2vCPU with a dedicated HF 1vCPU database server and a couple other HF 1vCPU instances for other websites that get trickle traffic.

    My forum (all XF) is not large, nor are my other forum(s) and websites. The "main" and largest forum is on 2vCPU/1vCPu (site/database) right now and displays this as an execution time:
    Code (Text):
    Width  Queries 30 Time 0.1729s Memory 9.31MB
    Would it perform better all on one bare metal instance of 4 core/8 threads & 32GB memory or am I better splitting them on separate instances?

    As it stands, my costs are about the same ($100) to host as is, but if I can get better performance out of the bare metal ($120), it would behoove me to change.

    (Cross-posted on XF forum)
     
  2. eva2000

    eva2000 Administrator Staff Member

    45,201
    10,280
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +15,934
    Local Time:
    10:43 PM
    Nginx 1.19.x
    MariaDB 5.5/10.x
    only way to know is test it out. Luckily both options you have hourly billing so that makes it much easier cost wise at least :)
     
  3. frm

    frm Member

    33
    4
    8
    Dec 17, 2018
    Ratings:
    +5
    Local Time:
    9:43 PM
    1.15.7
    10.3.11
    Is there perhaps a way to emulate simultaneous traffic (say 100 guests + me) browsing around in order to see if the execution times improve? Like create a new instance that is essentially a crawler, using CURL, to just go page after page and have 100 running at once?

    I know this is out of the realm of support here, but it doesn't quite seem as simple as copying and seeing if the execution time improves as it'd merely be myself on a test server unless I switched everything entirely over and then went back if there were no noticeable improvements from what I'm seeing.
     
  4. tininho

    tininho Premium Member Premium Member

    118
    26
    28
    May 22, 2019
    Nordic
    Ratings:
    +78
    Local Time:
    3:43 PM
    I did not find Vultr Bare Metal to have particularly good price/performance ratio (Xeon e3-1270v6) At least on performance side, you can have similar server with better price (Hetzner/OVH etc.)

    Vultr HF is still fastest VPS to my knowledge (at least for Wordpress), even Upcloud's Epyc seems to fall a little bit behind.

    I would not bet on your site becoming more responsive with Bare Metal (or improvement on TTFB or page load time). Maybe it can scale better, eva2000 has loads on load testing all over the forum.
     
  5. frm

    frm Member

    33
    4
    8
    Dec 17, 2018
    Ratings:
    +5
    Local Time:
    9:43 PM
    1.15.7
    10.3.11
    I've seen benchmarks for Vultr, as to why I joined (I believe through here, depositing $100 for a $100 match a minute ago), but only for VPS (high frequency vs regular vs Linode, etc.).

    Can't find anything on bare metal and which services would be best in that regard.

    I chose Epik as my domain registrar and their DNS is EPIK with 1-2ms DNS lookup (also a CDN). With them, my TTFB significantly increased alone. Just trying to squeeze every bit of speed out that I can, while keeping price about the same and concerned about security (all on one VPS [bare metal] or separate instances).
     
  6. eva2000

    eva2000 Administrator Staff Member

    45,201
    10,280
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +15,934
    Local Time:
    10:43 PM
    Nginx 1.19.x
    MariaDB 5.5/10.x
    moved your thread to VPS forum as it's more on topic there :)

    Did Vultr Bare Metal dedicated server benchmarks at Vultr - Vultr 60% discount & Free $100 Credit Test Drive Vultr Bare Metal Instances! including benchmarks for Nginx load balanacing versus Haproxy load balancing.

    Xenforo debug reported page times are mainly PHP processing. So like @tininho's WPPerformanceTester benchmarks, it's ultimately dependent on raw cpu clock speed, cpu generation (newer faster) and cpu allocation policies (if on VPS). Dedicated servers will usually be faster for PHP processing if clock speed, cpu generation are faster than VPS.

    systematically disable XF addons and re-measure your execution times to see which XF addon is causing higher processing time.

    Here's my XF 2.1 test forum with ~30 XF addons installed on 2 cpu core AMD Ryzen 3900X with 2GB ram and 50GB disk space OpenVZ VPS

    Index page
    Code (Text):
    Queries 15 Time 0.0573s Memory 3.03MB
    

    clicking on the time, will re-evaluate and breakdown the page load
    Code (Text):
    Page Time: 0.0249s
    Memory: 2.9193 MB (Peak: 3.0973 MB)
    Queries (16, time: 0.0025s, 10.0%)
    

    you can see majority of time is due to PHP processing 0.0249 - 0.0025 (mysql) = PHP time
     
  7. frm

    frm Member

    33
    4
    8
    Dec 17, 2018
    Ratings:
    +5
    Local Time:
    9:43 PM
    1.15.7
    10.3.11
    I was on the high frequency 2vCPU for files and 1vCPU for the database and these were 3 random samples for XF:
    Code (Text):
    Width  Queries 29 Time 0.1593s Memory 8.25MB
    Width  Queries 25 Time 0.1442s Memory 7.47MB
    Width  Queries 25 Time 0.1456s Memory 7.56MB
    I moved to 4vCPU dedicated cloud with the 1vCPU for the database and it only cut down on about 0.01-0.02 per run time with the same memory usage (didn't log it).

    The bare metal server (E3-1270v6 Processor 4 Cores - 8 Threads @ 3.8Ghz with 32768MB Memory [software SSD RAID]) blew both out of the water with 3 random results (note: the DB was moved to this localhost and not run on 1vCPU):
    Code (Text):
    Width  Queries 25 Time 0.0398s Memory 7.40MB
    Width  Queries 25 Time 0.0425s Memory 7.49MB
    Width  Queries 25 Time 0.0423s Memory 7.43MB
    I plan on shifting 4 other smaller XF forums to the server and I believe it'll still even outperform dedicated cloud. Getting someone like @eva2000 (a good sysadmin) on the server to tweak MySQL, Memcached, pagespeed, etc., could definitely enhance it even more, for sure, as I don't know which settings can get even a millisecond more out of it.

    And that concludes my test.

    Sticking with bare metal and object storage for redundant backups (as you can't have snapshots). :)