[Phpmyadmin-devel] 20 seconds lost in header_meta_style.inc.php

[Marc Delisle]

Garvin Hicking a écrit :
> Hi Marc!
> 
>> ok, using this require_once './libraries/header_http.inc.php'; $time_before =
>> microtime(true); require_once './libraries/header_meta_style.inc.php';
>> $time_after = microtime(true);
>> echo $time_after - $time_before;
>>
>> I only get .0006 seconds. No significant differences when OBgzip is true
>> or false. But still 20 seconds on my watch.
> 
> Okay, basically what I thought it would yield.
> 
>> Web server A (Apache 1.3.34) is on a different machine than MySQL server
>> B (MySQL 5.1.7, with 4400 dbs). When connecting Web server A to MySQL
>> server C (MySQL 5.0.18, 100 dbs) there is almost no delay. I'm using Firefox.
> 
> I do think that the time delay you experience comes from the server table
> cylcing...
> 
>> So, something else is eating time but it's difficult to find using
>> conventional tracing. I guess I'll have to use a debugger.
> 
> Absolutely, yes. I'd suggest using the XDebug2 Profiler, I think I remember
> you'Ve worked with that before?

Yes, to find that blowfish.php was eating many cycles, this is when we 
added mcrypt support. I don't have a machine nearby for this, it'll have 
to wait.

> 
>> P.S. what is HTTP chunking? :)
> 
> HTPP 1/1 supports HTTP chunking; this splits HTTP result data into multiple
> chunks of data, and even also to compress each of those chunks. This often
> speeds ups transfer, because the server can already deliver the first chunks of
> data while it is still chewing on the rest of the execution. I think though this
> only works with mod_php, not CGI's.
> 
> So in that case, if a chunk is 2kb in size, it will send the first 2048 bytes of
> your HTTP request; and then it might take the script to load several other files
> before the next 2kb of data are sent; this could be the reason why you thought
> the delay happened inside that meta file, while in "real time" the slow response
> is dealt with in some DB execution.

Thanks,
can't I force output with flush() ?

> 
> Best regards,
> Garvin
> 
>