[Page Previous 1, 2, 3, 4 Next]

Topic: PHP 5.6.4, 5.5.20, 5.4.36, 5.3.29 (x86/x64) available

[2]

Topic: PHP 5.6.4, 5.5.20, 5.4.36, 5.3.29 (x86/x64) available

[C0nw0nk]

Jan can i request you compile your builds with ssdeep

http://pecl.php.net/package/ssdeep

It is a extension i want to test with on windows i am working on preventing duplicate files and one site that currently uses this extension is www.virustotal.com

ssdeep :
384:1mDEF82t2udsptLuy3QCwiUDK68o7ZRH8vIDcSLNAafPdXpq1ym+6aFSonT6Ir0y:1mDE7tUtLPQC8K6t9KvDC9d3r0w3TMq

That would be a file identification key just like if you did md5_file(); or sha1_file();

So if we used the ssdeep extension we could check with ssdeep_fuzzy_hash_filename();

[gijs]

Why don't you make use of deduplication in Windows Server 2012 and newer?

It works on blok model so if a file is 99% identical it will only save the 1% difference.

[C0nw0nk]

Because when you run a site with a mysql database that points to files it wont work will it. You can remove the duplicate file but the path will now return a 404 error by your web server and the database will still contain the data. That would break the site basically and cause allot more problems you want a method that functions with your database and existing code.

You need to use a method that works with your code. This being my answer.

I currently use sha1_file(); But the longer the string identifyer the less likely you are to encounter another file with the same hash blueprint.

for example :

picture.jpg has a salted hash of "a1fb9d651cd328cb83b43e616189b7406ee6f889"

if i take picture.jpg rename it to picture1.jpg and reupload it the hash sum is still the same due to the contents of the file being the same this prevents people from uploading the same picture more than once with different file names.


If you want to see this code in action for yourself go to www.virustotal.com upload a file then rename the file and upload it again and it recognizes the same file because the hashes (basic blueprint) that can be used to identify a file are the same.

[gijs]

It doesn't change the path or anything like that it works on file system level, applications are not aware of it.

[C0nw0nk]

[gijs]

No, you don't understand.

It doesn't actually delete the duplicate files, it keeps a reference of every file and points to the data.

I suggest you read: http://blogs.technet.com/b/filecab/archive/2012/05/21/introduction-to-data-deduplication-in-windows-server-2012.aspx

[Jan-E]

Ok, this was a real challenge. It took me some 6 hours of work, but I got it running:

[C0nw0nk]

[gijs]

Yup, but you can't prevent the upload unless you check the filechecksum client side which isn't possible with PHP.

[C0nw0nk]

Well i use sha1_file(); to check uploads and i deal with allot of uploads everyday and these are video uploads upto 2gb in size and speed and performance is not effected. I even use Wincache with this and and process and convert all uploads into HD videos both MP4 and Webm format then i do the sha1_file(); on those and log their hashes to the database to prevent users downloading them and uploading it trying to be a pain or bypass the system.

[jimski]

Jan-E I have a question.

I haven't tested any of your latest builds that you just posted however almost every build of php from any source has the same problem of failing an apache bench test when doing phpinfo(). This includes official php builds from php.net

Let me be more specific.
When doing ab test on a localhost using a php file that contains only one line <?php phpinfo(); ?>
ab -n 10000 -c 100 -k http://localhost/phpinfo.php

This produces large number of failed requests on all versions of php except for php 5.3.x

Can you shine some light on this ?

[Jan-E]

I could not reproduce it with PHP 5.5 nts as mod_fcgid, but I can try to recompile without the modules that PHP 5.4+ has, but PHP 5.3 not: php_event, php_jsonc and php_strict.

php_strict might be a culprit, because it could not be loaded together with suhosin: http://pecl.php.net/package/strict

Which versions do you want me to compile first?

[Jan-E]

I recompiled the PHP 5.5/6 x86 builds (and the PHP 5.3/4 x64) builds without php_strict, php_jsonc/d and php_event. Could you try these? Otherwise try the previous versions: they are still online, just lower the number.

Starting the recompile of 5.4 x86 and 5.5/6 x64 right now.

[jimski]

[jimski]

[Jan-E]

On a Centos 6-server with PHP 5.3 I also had 0 failures. But with higher PHP-versions I ran into failures as well (1015 out of 10000). The output below is on a Centos 6 server with Apache 2.4.10, PHP 5.6.4 as php-fpm, using Zend OPcache.

So this seems a non OS-specific problem. It might even be that Nginx with the same PHP 5.6.4 as php-fpm runs into the same failures.
http://apaste.info/2du

Edit On a Centos 6 server with PHP 5.3.29 I had to run ApacheBench a couple of times before getting error-free. Probably I had to fill the OPcache:
http://apaste.info/YqC

Mod Note: please use apaste.info for long console outputs

[jimski]

I also tested in on Centos 6.5 and got similar results as you.
No problem with php 5.3.3 and php 5.3.28

However higher versions of php produced failures at the same rate on centos as on windows.

Unless we find a reason for this problem any benchmark test on higher versions of PHP will be probably meaningless if PHP can't reliably serve it's own info function without an error

[ng4win]

[Jan-E]

It is a cache issue (Edit: in PHP 5.3!). If I disable Zend OPcache it goes well the first time, with PHP 5.3 29 as PHP-FPM (and a concurrency level of 1020).

Could you try ApacheBench on a Nginx server?
ab -n 10000 -c 1020 -k http://127.0.0.1/phpinfo.php

As far as I know ApacheBench should be able to connect to Nginx as well.

http://apaste.info/D7m

Mod Note: please use apaste.info for long console outputs

[Jan-E]

The same on another Centos 6 server (with a faster processor) with PHP 5.6.4 as FPM, reports about 9500 failures the first time and around 10% (1000 out of 10000) every next time. It does not matter if I enable or disable Zend OPcache, the 10% failure ratio stays the same.