The server is my test server which is a P3-650 with a U160 SCSI drive. Hello World Benchmarks: Server Req/s -------------------------- Apache+Static 866.54 lighttpd+Static 262.41 Apache+PHP+APC 454.70 lighttpd+PHP+APC 168.19 From everything I have read, this should not be the case.
on 24.02.2006 17:49
on 24.02.2006 19:40
Brian Moon wrote: > The server is my test server which is a P3-650 with a U160 SCSI drive. > > Hello World Benchmarks: > > Server Req/s > -------------------------- > Apache+Static 866.54 > lighttpd+Static 262.41 > > Apache+PHP+APC 454.70 > lighttpd+PHP+APC 168.19 > > > From everything I have read, this should not be the case. Two Things come to my mind when i see posts like this : First : For sure you're doing something terribly wrong. But are you really expecting other answers? Second : I don't think anybody could help you with that without any Information about your setup and your Test environment. A little success - story : Site : about 250-300 Mio PageImpressions per month 3 Servers each: Hardware : AMD3000+ @1Gb Ram + 160GbSata Software : Debian apache 1.33 heavily tuned with php4.4.2 + eaccellerator0.9.4rc1 balanced with mod_backhand + wrrd - Mysql 4.1 master/slave replicated on all 3 Servers as backend Caching : fragments where cached by Ea'shm Load with apache between 10 - 60. Then switched to lighttpd and Fastcgi -> Load drops immediately below 10 and is now betwenn 0,X - 7 . The Content of the site is some kind of Party Community with lots of pictures and a lot of php for all kind of Community functions such as messages and so on ( ~ 70 messages/s ) about 25000 pictures until now.
on 25.02.2006 00:08
Because I am new to lighty, I did not want to assume what data one would need to answer my questions. I can tell you that both apache and lighty were vanilla installs with little to no changes in the default configs. > Load with apache between 10 - 60. > > Then switched to lighttpd and Fastcgi -> Load drops immediately below 10 > and is now betwenn 0,X - 7 . I don't care about load, I care about speed. If the load is .09 but I only get 100 req/s, it does me no good. And, in fact, the load was not high at all during these tests. But, neither was the performance. I swear I am not trolling. I want in on the secret to making it work.
on 25.02.2006 00:23
Brian Moon wrote: > Hello World Benchmarks: please post full ApacheBench output (if this is ApacheBench) for these tests, for static use any static file, for dynamic please try to use some _real_ application URL - even better with SQL used, not just phpinfo() or Hello World etc. it will be a good start for further research Cheers, --- Albatros Vep Taloha http://host8.biz
on 25.02.2006 21:07
Brian Moon wrote: > > I don't care about load, I care about speed. If the load is .09 but I > only get 100 req/s, it does me no good. And, in fact, the load was not > high at all during these tests. But, neither was the performance. > I'd like you to know that each page consists of min. 15 files (pictures, css + 1 php) so if you'd like to have plain hits/s (as what req/s are) we can calculate 250000000*15/30/24/3600 ~ 482,25req/s per Server (remember this value is on average) To test by myself i went down and did it the same as you did (just for the sake of it) lighttpd 1.4.10, php5.1 + eaccellerator (no i did not remove unnecessary modules) ab -n10000 -c600 http://staging.XYZ.de/hello.php Server Software: lighttpd/1.4.10 Server Hostname: staging.XYZ.de Server Port: 80 Document Path: /hello.php Document Length: 11 bytes Concurrency Level: 600 Time taken for tests: 10.928466 seconds Complete requests: 10000 Failed requests: 0 Write errors: 0 Total transferred: 1561716 bytes HTML transferred: 110121 bytes Requests per second: 915.04 [#/sec] (mean) Time per request: 655.708 [ms] (mean) Time per request: 1.093 [ms] (mean, across all concurrent requests) Transfer rate: 139.54 [Kbytes/sec] received with -k Server Software: lighttpd/1.4.10 Server Hostname: staging.XYZ.de Server Port: 80 Document Path: /hello.php Document Length: 11 bytes Concurrency Level: 600 Time taken for tests: 9.436407 seconds Complete requests: 10000 Failed requests: 0 Write errors: 0 Keep-Alive requests: 10000 Total transferred: 1610322 bytes HTML transferred: 110022 bytes Requests per second: 1059.73 [#/sec] (mean) Time per request: 566.184 [ms] (mean) Time per request: 0.944 [ms] (mean, across all concurrent requests) Transfer rate: 166.59 [Kbytes/sec] received Server : Sempron64@3000+ 1GB Ram 100Mbit (linux 2.6.14.2) Client : 2*AMD2400+ 1GB Ram 100Mbit (Linux 2.6.15.4) Between Server and Client where 9 Hops (made through the internet) and latency 0f 8.92 ms (PING) Some Important Parts of lighttpd config : server.event-handler = "linux-sysepoll" server.max-fds = 4096 server.max-keep-alive-requests = 100 server.max-keep-alive-idle = 6 server.max-read-idle = 30 server.max-write-idle = 120 fastcgi : "socket" => "/tmp/php-fastcgi.socket", "bin-path" => "/usr/bin/php5-cgi", "check-local" => "enable", "max-procs" => 20, "bin-environment" => ( "PHP_FCGI_CHILDREN" => "15", "PHP_FCGI_MAX_REQUESTS" => "10000" ), "broken-scriptfilename" => "enable", "bin-copy-environment" => ( "PATH", "SHELL", "USER" ) no access.log =============apache 2.0.54======================== some words about apache2.0.54 config (most of you know that apache2-mpm-worker isn't used together with php because gd is not threadsafe (afaik) ) so i used mpm-prefork + same php version & ini (PHP 5.1.2-1.dotdeb.2) but as apache-module for sure KeepAlive On MaxKeepAliveRequests 100 KeepAliveTimeout 15 <IfModule prefork.c> StartServers 50 MinSpareServers 10 MaxSpareServers 30 MaxClients 500 MaxRequestsPerChild 10000 </IfModule> Maybe we can tweak another 100/reqs more per second with some better values these seemed to match this benchmark best (in my eyes) again : ab -n10000 -c600 http://staging.XYZ.de/hello.php Server Software: Apache/2.0.54 Server Hostname: staging.XYZ.de Server Port: 80 Document Path: /hello.php Document Length: 11 bytes Concurrency Level: 600 Time taken for tests: 36.823996 seconds Complete requests: 10000 Failed requests: 0 Write errors: 0 Total transferred: 1730000 bytes HTML transferred: 110000 bytes Requests per second: 271.56 [#/sec] (mean) Time per request: 2209.440 [ms] (mean) Time per request: 3.682 [ms] (mean, across all concurrent requests) Transfer rate: 45.87 [Kbytes/sec] received Connection Times (ms) min mean[+/-sd] median max Connect: 8 538 2217.0 9 21013 Processing: 11 709 1825.6 154 27438 Waiting: 9 707 1825.6 152 27437 Total: 19 1247 3140.6 167 33505 -k Server Software: Apache/2.0.54 Server Hostname: staging.XYZ.de Server Port: 80 Document Path: /hello.php Document Length: 11 bytes Concurrency Level: 600 Time taken for tests: 8.782287 seconds Complete requests: 10000 Failed requests: 0 Write errors: 0 Keep-Alive requests: 9911 Total transferred: 2096224 bytes HTML transferred: 110011 bytes Requests per second: 1138.66 [#/sec] (mean) Time per request: 526.937 [ms] (mean) Time per request: 0.878 [ms] (mean, across all concurrent requests) Transfer rate: 233.08 [Kbytes/sec] received => Hope i made no mistake and (i must admit) i was quite impressed about apache2 keepalive performance (i didn't expect something like this)
on 26.02.2006 17:55
> please post full ApacheBench output (if this is ApacheBench) > for these tests, for static use any static file, for dynamic > please try to use some _real_ application URL - even better > with SQL used, not just phpinfo() or Hello World etc. Well, I am hesitant to use database activity in these tests the goal for this application will not use any database connectivity. IMO, a hello world test is the proper baseline to use. Can light get faster with a "real" application than it is for Hello World? The primary use for the application will be to A) retrieve content from memcached and return it to the client and B) to serve static images. I did do another test that involved memcached. I don't have verbose data for it, but here is the summary: Memcache 100 gets: Platform Req/s ---------------------- Apache+PHP+APC 62.40 lighttpd+PHP+APC 55.14 As you can see, Apache was still faster in that test. I did do more verbose hello world and static file tests. Rather than post the massive text here, I have put them here: http://dev.phorum.org/lighty.txt http://dev.phorum.org/apache.txt http://dev.phorum.org/sysinfo.txt http://dev.phorum.org/lighttpd.conf I would really love to make this application go faster with lighty. I just can't figure out what I am doing wrong.
on 26.02.2006 20:08
If you want to run ab with -c600 and not -c60, then:
1. turn off debugging (!)
2. test && check lighty error log
3. set server.max-fds = 8192
4. test && check lighty error log
5. set some nice values for:
server.max-keep-alive-requests = value
server.max-keep-alive-idle = value
server.max-read-idle = value
server.max-write-idle = value
6. test && check lighty error log
now lighty should fly :-)
HTH
---
Albatros Vep Taloha
http://host8.biz
on 27.02.2006 01:41
> 1. turn off debugging (!) OMG, why would this be turned on in the example config that comes with the server? > now lighty should fly :-) Server Req/s -------------------------- Apache+Static 866.54 lighttpd+Static 1874.96 Apache+PHP+APC 454.70 lighttpd+PHP+APC 376.31 Full output: http://dev.phorum.org/lighty2.txt Its definitely better on static pages, but still slower on PHP. But, not by much. I am happy I was able to produce numbers closer to what I expected. FYI, that is with 200 PHP fcgi children. Apache has MaxClients at 150. So, lastly, I decided to see if there was an advantage when outputting lots of data. Our web pages can have as much as 100k of html in them. Server Req/s -------------------------- Apache+PHP+APC 90.64 lighttpd+PHP+APC 60.59 Full output: http://dev.phorum.org/bigoutput.txt Alas, that did not turn up any advantage either. So, for static files, lighty is over twice as fast.
on 27.02.2006 10:53
Please try with fastcgi.server via socket, not host/port HTH --- Albatros Vep Taloha http://host8.biz
on 27.02.2006 15:08
Albatros Vep Taloha wrote: > > Please try with fastcgi.server via socket, not host/port Slowly but surely we are creating what should be, IMHO, the default config. Server Req/s -------------------------- Apache+PHP+APC 444.29 lighttpd+PHP+APC 1022.45 I updated http://dev.phorum.org/lighty2.txt I need one more magic bullet for big output though. Server Req/s -------------------------- Apache+PHP+APC 90.64 lighttpd+PHP+APC 44.80 I updated http://dev.phorum.org/bigoutput.txt Any more config tricks up your sleeve for making output faster? I ran that last test 4 times and could not make it faster. You will note, it actually got slower than the test where I used tcp for fcgi.
on 27.02.2006 18:01
It would be interesting to see your fastcgi.server = (...) lighty config file part --- Albatros Vep Taloha http://www.host8.biz
on 27.02.2006 18:05
config at http://dev.phorum.org/lighttpd.conf php started like: /usr/local/bin/spawn-fcgi -s /tmp/fgci-php.sock -f /usr/local/bin/php -u nobody -g nobody -C 200
on 27.02.2006 18:31
Since you are using fast-cgi on localhost,
and probably new PHP version, it is better
to not start separate PHP, so it can look like:
fastcgi.server = ( ".php" => ((
"bin-path" => "/usr/bin/php-cgi -c /usr/etc/php.ini",
"socket" => "/tmp/php-fastcgi.sock-3-" + var.PID,
"min-procs" => 1,
"max-procs" => 3,
"max-load-per-proc" => 3,
"idle-timeout" => 60,
"bin-environment" => (
"PHP_FCGI_CHILDREN" => "50",
"PHP_FCGI_MAX_REQUESTS" => "10000"
),
"bin-copy-environment" => (
"PATH", "SHELL", "USER"
),
"broken-scriptfilename" => "enable"
)))
now you can play with better values for 'max-procs'
and 'PHP_FCGI_CHILDREN' above (+/-)
try also with lighty config values as below:
server.max-keep-alive-requests = 5
server.max-keep-alive-idle = 5
server.max-read-idle = 30
server.max-write-idle = 120
HTH
---
Albatros Vep Taloha
http://www.host8.biz
