[Page Previous 1, 2, 3, 4, 5, 6]

Topic: Nginx vs Other Webservers on WINDOWS ONLY

[6]

Topic: Nginx vs Other Webservers on WINDOWS ONLY

[C0nw0nk]

Thanks because its windows i read that mmap should be set like the following not as a file path ?

"for win32, xcache.mmap_path=anonymous map name, not file path"

[James Blond]

I know you all love this topic, but it is no longer the X VS Y topic, but how to configure nginx. Please feel free to open a new topic in the Other Software section. Thanks!

[C0nw0nk]

Sorry James i created the following topic to continue on my discussion about caching in regards to PHP and Joomla.

http://www.apachelounge.com/viewtopic.php?p=27992#27992

[C0nw0nk]

@ng4win i found a bug with what i recon is fastcgi in your version.

Here is how i know it can only be fastcgi.

PHP outputs a page load speed. (Page load speeds on all pages are always less than 1 second)

Mysql i have enabled slow query logging that logs queries that take more than 2 seconds. (There are no queries that take more than 2 seconds)

Now very often pages take 30 seconds or more to load but yet if i am to access a page through nginx that is not php. (Html) it loads instantly all the time.

Only pages that get delieverd by fastcgi are taking sometimes up to a miniute to load but yet my php execution "Time to create page: 0.125 seconds" Something is very wrong.

I use the following firefox extension https://addons.mozilla.org/en-US/firefox/addon/extended-statusbar/?src=userprofile

"Time to create page: 0.27 seconds" and firefox output was "33 seconds", It can only be fastcgi since all other requests that do not hit fastcgi mp4, html, static requests all work fine and fast.

[ng4win]

I doubt its a bug since control is passed to php-cgi and then its a waiting game for when php returns the results, could be anything within php/mysql, anyway I need a working config to reproduce this.

[C0nw0nk]

Do you mind if i send you a email ? http://www.ecsystems.nl/contact.html

I'd be willing to give you RDP access.

[ng4win]

We only use teamviewer and such support is considered commercial support. Post a new topic like the mods noted yesterday, or on the nginx forums with the basic fastcgi/php.ini config and a good description what is happening and what you think should happen. Also remove wincache for the time being.

[C0nw0nk]

Fair enough i posted it over at the nginx mailing forum will wait for replies there http://forum.nginx.org/read.php?2,250964

[C0nw0nk]

In the battle of Nginx vs "The World" who has written or has a good script / command line (.cmd/.bat) file for me to do some testing and benchmarking with

[jimski]

Hi ng4win, I have to say that the documentation on your website needs a lot of work to be user friendly.

For example there is no explanation what the following files do:
ntrights.exe
srvany.exe
wget.exe
sleep.exe

The one line "Used by service creation and installation scripts" is insufficient. As a sysadmin I'm always cautious of executables with no explanation.

Also installation steps are convoluted and using vb scripts to unzip a file seems to be overkill.

Of course an experienced sysadmin can get it to work (after some hair pulling) but this won't get your work a wide acceptance which you most likely deserve.

[ng4win]

[C0nw0nk]

[ng4win]

Tnx but powershell is not present by default, you need to install it which is what we try to avoid, nothing is needed to get going. This vb stuff runs on anything.

[jimski]

[ng4win]

If mod_php is an internal process compared to nginx/php which uses a tcp port then yes mod_php will be faster inherent to design. OTOH nginx/Lua will blow apache/Lua out of the water.

[jimski]

ng4win, have you tested nginx for windows serving the phpinfo() page on windows with apache bench or any other performance software? BTW only php 5.3.x will serve phpinfo() without errors when using apache bench.

How many requests per second can you serve when using phpinfo() as a test page. Also let us know on what hardware.

In the past we have used this testing methodology for testing Apache with FastCGI:

TESTING METHODOLOGY:

1. Disable any Caching in php and on webserver.

2. Use this php file. Copy the line below and save it as phpinfo.php. It is just one line but gives apache and php good beating.

<?php phpinfo(); ?>

3. Set FcgidMaxProcesses 1
FcgidMaxProcessesPerClass 1
FcgidProcessLifeTime 0
FcgidMaxRequestsPerProcess 0

4. Run Apache Bench with these parameters and save results
ab -n 10000 -c 1 http://127.0.0.1:80/phpinfo.php

5. Then set FcgidMaxProcesses 10
FcgidMaxProcessesPerClass 10
FcgidMinProcessesPerClass 10

6. Restart Apache

7. Run Apache Bench with these parameters just to spawn the processes but ignore the results
ab -n 10000 -c 100 http://127.0.0.1:80/phpinfo.php

8. Run Apache Bench again with these parameters and save your results
ab -n 10000 -c 100 http://127.0.0.1:80/phpinfo.php

9. Post your results using these 4 lines only:
Time taken for tests: xxx
Complete requests: xxx
Failed requests: xxx
Requests per second: xxx

[ng4win]

It's a nice idea but this will turn out to be a 'my d*ck is bigger then yours' contest...

You can't compare such results, for example everything here is virtual running on 1 to 16 clusters depending where your request comes in.

Using real hardware doesn't mean much either unless they are absolutely the same including the network hardware and wire-ring.

And what about sustained throughput testing? the same issues and also the difference in design, actual available bandwidth (iperf -w 256k anyone?), etc.

What about blocking vs. non-blocking configurations? I could easily embed php inside nginx for some simple php stuff but fully non-blocking and hand you a "Requests per second: 10000.00 [#/sec]" result which is meaningless due to the design differences.

2 weeks ago I saw 48 workers on 8 cores, they were complaining it slowed down too much.... duh.... I offered them their subscription money back, I wasn't even going to bother explaining why.

Yet on our OTA cluster we test with 32 workers also on 8 cores which are hyper-threaded with 2 pipes from the main board to each cpu (only then hyper-threading makes any point in using) which effectively makes 2 workers/core, with a low prio/worker and ramdrives for the content.

To do a close enough compare even with Lua would give results you can't really compare.

[jimski]