Join the community today
Become a Member

WebPerf [Dropbox] Optimizing web servers for high throughput and low latency

Discussion in 'All Internet & Web Performance News' started by eva2000, Sep 7, 2017.

Tags:
  1. eva2000

    eva2000 Administrator Staff Member

    36,054
    7,910
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +12,192
    Local Time:
    7:27 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    Nice article by folks at Dropbox https://blogs.dropbox.com/tech/2017/09/optimizing-web-servers-for-high-throughput-and-low-latency/

     
    • Winner Winner x 1
  2. rdan

    rdan Premium Member Premium Member

    4,255
    1,034
    113
    May 25, 2014
    Ratings:
    +1,486
    Local Time:
    5:27 AM
    Mainline
    10.2
  3. pamamolf

    pamamolf Well-Known Member

    3,113
    295
    83
    May 31, 2014
    Ratings:
    +530
    Local Time:
    12:27 AM
    Nginx-1.13.x
    MariaDB 10.1.x
    Don't know if enabling this may help attackers to use such packets to attack or not?

    #recommended for hosts with jumbo frames enabled
    net.ipv4.tcp_mtu_probing=1
     
  4. eva2000

    eva2000 Administrator Staff Member

    36,054
    7,910
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +12,192
    Local Time:
    7:27 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    Alot of the advanced optimisations outlined in that article would need a good understanding of it's implications for specific usage scenarios so like the article says, test for your circumstances and usage and see if it works for you.
     
    • Informative Informative x 1
  5. bassie

    bassie Active Member

    939
    220
    43
    Apr 29, 2016
    Ratings:
    +664
    Local Time:
    11:27 PM
    I ran into this Dropbox article today.
    In addition to the above.

    Most tweaks from the above article will only really take effect if you have millions of users (Dropbox, Facebook, Google et al). Then every 0.1% counts.

    The question is whether it really saves on relatively small websites that we work with.

    I recently started using the less is more principle again and said goodbye to most patches and modules. Like Nginx Brotli, Google BBR, TCP Fast Open, Cloudflare HPACK and Cloudflare SPDY, Nginx Brotli, Chacha if Android prefer etc etc.

    It does not matter at all in terms of search engine visitor potential and analytics confirm that.

    It also simply takes too much time to adjust everything time and time again with a new Nginx version or other software component. And then I have to talk about the test work so everything is ultra stable.

    And all of that for not even 1% profit?

    You can spend your time only once and there is better spending.
     
    • Informative Informative x 1
  6. eva2000

    eva2000 Administrator Staff Member

    36,054
    7,910
    113
    May 24, 2014
    Brisbane, Australia
    Ratings:
    +12,192
    Local Time:
    7:27 AM
    Nginx 1.15.x
    MariaDB 5.5/10.x
    Yeah that's why I automate most of these and have individual enable/disable variables to control each patch + have nginx version checks to determine if a patch can be used i.e. Cloudflare HPACK Full Encoding patch isn't enabled if Nginx 1.13.10+ or higher are detected. And most of these patches are disabled by default in Centmin Mod, unless user decides to enable them.

    But yes it does take more time figuring it out initially or when a Nginx version breaks compatibility with them. But it's partially the reason why I created Centmin Mod LEMP stack the way it is with PHP and Nginx source compiled, as I want to have the optional ability to try out the latest bleeding edge web tech on offer and not be constrained or limited as I would be RPM package solutions like other LEMP/LAMP stacks.
     
    • Like Like x 1
..