Join the community today
Register Now

1 large VPS or multiple smaller?

Discussion in 'Virtual Private Server (VPS) hosting' started by Daniel J. Lewis, Sep 13, 2016.

  1. Daniel J. Lewis

    Daniel J. Lewis Award-winning podcaster and consultant

    77
    8
    8
    Oct 20, 2014
    Ratings:
    +11
    Local Time:
    1:53 PM
    1.8.0
    5.6
    I'm debating this issue right now. I have a WordPress Multisite with a 2+ GB database, 12 high-traffic sites running on its own 4 GB / 4 CPU VPS on Vultr (an older instance before they slowed the CPUs). I also have multiple single low-traffic WP sites on a second 2 GB / 2 CPU VPS.

    Is this separate-server setup a good idea, or would I be better with a single, upgraded server for everything?
     
  2. bassie

    bassie Active Member

    572
    124
    43
    Apr 29, 2016
    Ratings:
    +376
    Local Time:
    7:53 PM
    Separate servers if you ask me.
    So your whole environment isn't down in case of the 1 server setup with it's single point of failure.
     
    • Like Like x 1
    • Informative Informative x 1
  3. eva2000

    eva2000 Administrator Staff Member

    30,966
    6,919
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +10,422
    Local Time:
    4:53 AM
    Nginx 1.13.x
    MariaDB 5.5
    Really is just a judgment call. Usually, my rationale for separate servers is just so that if one site takes off traffic wise, it's just easier to scale up or move servers for that one site without other sites on the server needing to take account for. Automated backups also is easier to manage disk space requirements. It's also why i prefer subdomain over directory for sites, makes scaling a site much easier i.e. forum.domain.com versus domain.com/forum. If forums grow, just move the forums itself instead of entire site. Of course if you 100% know that site isn't going to grow traffic wise much, I'd probably group them together on one server.
     
    • Agree Agree x 1
  4. Daniel J. Lewis

    Daniel J. Lewis Award-winning podcaster and consultant

    77
    8
    8
    Oct 20, 2014
    Ratings:
    +11
    Local Time:
    1:53 PM
    1.8.0
    5.6
    I'm reviving this old thread because I'm considering this option again.

    My servers on Vultr are still as above (WP Multisite on a 4-core, 4 GB RAM server; all other sites on a 2-core, 4 GB RAM server), and I have the older CPUs still at 3.6 GHz.

    Downtime or spike possibilities aside, I'm thinking about performance. Would it be better to move everything onto a single, 4-core/8 GB RAM server (Vultr will let me upgrade that 4-core server to double the RAM)? This means that all my sites would have access to 8 GB of RAM instead of only the 4.

    Or, do you think I should keep what I have?
     
  5. ArisC

    ArisC Member

    46
    14
    8
    Jun 1, 2017
    Ratings:
    +21
    Local Time:
    8:53 PM
    Nginx 1.13.6
    MariaDB 10.1.28
    Load Balancing. You can Achieve 100% Uptime and prevent spikes
     
  6. Daniel J. Lewis

    Daniel J. Lewis Award-winning podcaster and consultant

    77
    8
    8
    Oct 20, 2014
    Ratings:
    +11
    Local Time:
    1:53 PM
    1.8.0
    5.6
    That doesn't really answer my question, and I have no idea how to do load-balancing.
     
  7. jscott

    jscott Member

    94
    14
    8
    Aug 13, 2015
    Ratings:
    +31
    Local Time:
    1:53 PM
    Daniel,

    <bad accounting joke>
    This is kind of like asking a CPA what is the better option... the answer is... It depends!
    </bad accounting joke>

    One one hand....
    I tend to think of it as a basket of eggs. If one server chokes, you have a dozen large clients / users all yelling at you at one time.

    Myself, I would probably put one or two large and a few small sites on several servers.

    This way, if there is some problem with hardware or with a system configuration change, all of your clients are not mad at you.

    On the other hand....
    If these are just hobby sites they may not be bringing in any money. So money may be tight enough to make multiple servers hard to justify.

    So... as the accounting joke goes... It depends....

    -John
     
    • Like Like x 1
  8. eva2000

    eva2000 Administrator Staff Member

    30,966
    6,919
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +10,422
    Local Time:
    4:53 AM
    Nginx 1.13.x
    MariaDB 5.5
    Know my stance in post #3 above :) FYI, sites don't get access to all memory available on server, they only access what is required and/or what you tell them to access. Just merging all sites on a larger memory based server doesn't mean that all sites get access to all that extra memory.
     
  9. bassie

    bassie Active Member

    572
    124
    43
    Apr 29, 2016
    Ratings:
    +376
    Local Time:
    7:53 PM
    It is difficult to determine performance as you are on a shared environment.
    In other words the new single 4-core/8 GB RAM server could be significantly slower If you have busy neighbors.

    In addition, the CPU is often more important than memory for busy sites.
    Apart from the fact that new Vultr CPU cores are capped (no more 3.6 GHz), You will also have much less CPU cores, for the whole, new combined environment (now 4 +2 (then 4 - 2 GHz)).
    This is a considerable deterioration.
     
    Last edited: Aug 31, 2017
  10. Daniel J. Lewis

    Daniel J. Lewis Award-winning podcaster and consultant

    77
    8
    8
    Oct 20, 2014
    Ratings:
    +11
    Local Time:
    1:53 PM
    1.8.0
    5.6
    I'm still on the old Vultr instances and when I upgrade, I get to keep the CPU speeds (3.6 GHz).

    But it's sounding like maybe it's best for me to keep the two separate servers.
     
  11. bassie

    bassie Active Member

    572
    124
    43
    Apr 29, 2016
    Ratings:
    +376
    Local Time:
    7:53 PM
    You still lose 2 cores.
    I assume you monitor everything. What about your current usage?
     
  12. Daniel J. Lewis

    Daniel J. Lewis Award-winning podcaster and consultant

    77
    8
    8
    Oct 20, 2014
    Ratings:
    +11
    Local Time:
    1:53 PM
    1.8.0
    5.6
    Well, I still feel like an amateur in running my own servers. I sometimes monitor CPU usage with htop -d 2 and it spikes only when I initiate some intense process (like preloading cache). My memory is almost always in high usage because of many instances of mysqld all somehow using 52.6% of memory on the 4 core / 4 GB server (with the massive database and WordPress Multisite), and 6.1% on the 2 core / 4 GB server running multiple single-site WordPress installs.
     
  13. bassie

    bassie Active Member

    572
    124
    43
    Apr 29, 2016
    Ratings:
    +376
    Local Time:
    7:53 PM
    It's never too late.
    Htop is a system-monitor process-viewer.
    Just a moment of recording, no monitoring.

    You should start monitoring your environment for some time.
    (My preference would be a couple of weeks)

    Then analyze your data and you will soon see what you need.

    Without such data, an analysis is impossible and you never know exactly what you need. Or if there are bottlenecks, your server is overkill or oversold etc etc.

    You could start with nodequery, nixstats or paid newrelic.
     
  14. eva2000

    eva2000 Administrator Staff Member

    30,966
    6,919
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +10,422
    Local Time:
    4:53 AM
    Nginx 1.13.x
    MariaDB 5.5
    might want to read Help! Linux ate my RAM!
     
    • Like Like x 1
  15. Daniel J. Lewis

    Daniel J. Lewis Award-winning podcaster and consultant

    77
    8
    8
    Oct 20, 2014
    Ratings:
    +11
    Local Time:
    1:53 PM
    1.8.0
    5.6
    Cool. I knew that's how it worked, but I guess I was concerned since so many MySQL processes were marked as using a lot of RAM. But if that's not an issue, I'll not be concerned about that.

    I installed some monitoring tools, but I also looked at my Vultr Dashboard.

    This is server NDL, which is the 4-core / 4 GB RAM running WordPress Multisite, bbPress, BuddyPress, and WooCommerce:

    Screenshot 2017-08-31 16.18.13.png

    This is server DJL, which is the 2-core / 4 GB RAM running multiple single-site WordPress installations:

    Screenshot 2017-08-31 16.18.17.png
    (Those last couple spikes are cache preloading.)