RE: [PECL-DEV] Making it easier to leverage APC
| From: | Andi Gutmans | Date: | Sun, 03 Dec 2006 07:15:49 +0000 |
| Subject: | RE: [PECL-DEV] Making it easier to leverage APC | ||
| References: | 1 | Groups: | php.pecl.dev |
| Request: | Send a blank email to pecl-dev+get-4381@lists.php.net to get a copy of this message | ||
Faster is always nice but the biggest and most noticable performance gain
happens. I believe that the performance loss is significant in most
real-world apps.
I just pointed it out because PHP users who don't know internals who hear
this kind of stuff think it's a real issue and start worrying about it, when
they really don't have to.
Andi
> -----Original Message-----
> From: Rasmus Lerdorf [mailto:rasmus@lerdorf.com]
> Sent: Saturday, December 02, 2006 8:38 PM
> To: Andi Gutmans
> Cc: 'Marcus Boerger'; 'Lukas Kahwe Smith'; pecl-dev@lists.php.net
> Subject: Re: [PECL-DEV] Making it easier to leverage APC
>
> To clarify, of course conditionally included files get
> compiled and cached. The issue is not the included files but
> the resulting conditionally defined classes and functions
> needing to be redefined on every request. Whether that is
> significant or not comes down to the specifics of the
> situation, but there is no doubt that it is slower. It comes
> down to a NOP vs. a FETCH_CLASS, for example and the NOP is
> obviously way faster.
>
> -Rasmus
>
> Andi Gutmans wrote:
> > I don't see a reason for byte code caches not to work with
> > __autoload() and conditional classes already today. I don't know of
> > the internals of APC but
> > __autoload() should not really be an issue (it's just another
> > include()) and I'm not quite clear why Rasmus thinks that
> doesn't work.
> >
> > As far as conditional class declerations are concerned,
> it's also not
> > a huge issue. It might not be quite as fast as ones in the
> main block
> > but you still get the majority of the bytecode cache
> benefit which is
> > that you don't need to re-read the file, do lexical
> analysis and run a parser on it the code...
> >
> > Andi
>