How would you host the most websites?

BlueVM

Alright so let's get into some nitty gritty stuff since I like talking about theoretical projects and how they'd run.

Assuming you had 1 GB of RAM and wanted to host an ever growing number of websites to the point that they were as optimized as possible... how would you do it?

Limitations: I'm aware there is no such thing as unlimited storage or bandwidth (no way? really?) - That isn't a factor... pure and simple RAM/CPU usage to a minimum...

Derek

Nginx + PHP 5.3.10 + APC Enabled as opcache and for caching fequently used queries. MySQL tuned to the point of no more tuning.

Kairus

@Derek said: Nginx + PHP 5.3.10 + APC Enabled as opcache and for caching fequently used queries. MySQL tuned to the point of no more tuning.

This.

natestamm

@BlueVM see above ...can also ask browser to respond to gzip compression.

Adam

@Derek said: Nginx + PHP 5.3.10 + APC Enabled as opcache and for caching fequently used queries. MySQL tuned to the point of no more tuning.

This plus Supervisord and Beanstalkd

Maounique

@natestamm said: see above ...can also ask browser to respond to gzip compression.

I am not sure using gzip will be easier on the CPU or RAM.
M

ElliotJ

Pre-Gzip everything, and then force the web server to send the gzipped versions.
I'm sure nginx supports this.

You could throw CloudFlare onto it as well, reduce the requests a bit.

OneTwo

1 main web server: lighttpd (gzip) + fastcgi + php-xcache
1 mysql server: mysql only (to cache shit)
2 front-end servers: nginx (gzip)

iptables rules to the backend servers for security and shit. that would work for many req/s.

i'm running a site with static content which have an average of 8 req/s since I restarted the webserver runs awesome on a single 256MB Ram vps with lighttpd (gzip). the only problem is the bandwidth.

titanicsaled

How about trying varnish cache?

OneTwo

Getting a Xeon E7 on a gigabit connection would work too.

Steve81

squid in reverse proxy mode (without disk cache) + cherokee + php-fpm on bsd sockets + apc

Works flawless on a 128mb xen vps.

lbft

nginx's FastCGI cache options can provide a significant boost for non-static content if you don't have a proxycache like Squid or Varnish in front of it; it turns serving a dynamic page into serving a static page from disk.

It's simple to set up and will make more efficient pretty much any request that can be safely cached for a set period of time (minutes/hours). Speedy page loading is one of the factors Google uses in ranking search results too, for added benefit.

marrco

@lbft that's interesting, care to share some config and explain better?

FLVPS

I'm messing around with this idea on cPanel with 1024mb of RAM. I'll keep you updated on how it goes.

pro

nginx + unicorn for ruby/rails

nginx for GoLang code. I also use nginx + thttpd for GoLang when I need to run in CGI.

lighttpd for ikiwiki / perl (CGI) / CGI

djvdorp

@Steve81 said: squid in reverse proxy mode (without disk cache) + cherokee + php-fpm on bsd sockets + apc

interesting config, might wanna share the details :P ?

Steve81

@djvdorp said: interesting config, might wanna share the details :P ?

There aren't a lot of details, is a quite simple configuration.

However:

• cherokee listening on 127.0.0.1:80 with gzip/deflate compression enabled (except for php), compression level set to max.
• Squid listening on IP:80
  • Squid config:

    acl all src all
    acl manager proto cache_object
    acl purge method PURGE
    acl CONNECT method CONNECT
    
    # "default" is, on my vps, a vhost that return a 401 Unauthorized;
    # if someone doesn't know what is the hostname of my website,
    # he doesn't need to see it
    http_port IP:80 accel defaultsite=default vhost
    
    # Cache_mem depends of the avaliable ram on the machine
    # I use 8MB on a 64mb vps
    cache_mem 8 MB
    
    # No disk cache
    cache_dir null /var/spool/squid
    
    access_log none
    cache_log none
    cache_store_log none
    icp_port 0
    udp_incoming_address 127.0.0.1
    hosts_file /etc/hosts
    coredump_dir /var/spool/squid
    
    # cherokee is listening on 127.0.0.1:80
    cache_peer 127.0.0.1 parent 80 0 no-query no-digest no-netdb-exchange originserver
    name=cherokee login=PASS
    
    # All requests are sent to the local webserver
    # Just to avoid to write a list with all available vhosts
    cache_peer_access cherokee allow all
    
    # I don't want to expose that I cache throught Squid
    header_access Via deny all
    
    # Needed to allow cherokee to replace REMOTE_ADDR with the real REMOTE_ADDR ip
    forwarded_for on
    header_access X-Forwarded-For allow all
    
    http_access deny manager all
    http_access deny purge all
    http_access deny CONNECT all
    
    # All requests are sent to the local webserver
    http_access allow all
    

djvdorp

might be a dumb question, but are there advantages from Cherokee over Nginx for example?

Steve81

@djvdorp said: might be a dumb question, but are there advantages from Cherokee over Nginx for example?

I found some problems with php-fpm on unix sockets & nginx. Apart from that, cherokee have a web interface to easily configure the webserver.