Newsgroups: php.internals
Path: news.php.net
Xref: news.php.net php.internals:66623
Mailing-List: contact internals-help@lists.php.net; run by ezmlm
Received-SPF: error (pb1.pair.com: domain garfieldtech.com from 66.111.4.25 cause and error)
Message-ID: <51420D30.8060005@garfieldtech.com>
Date: Thu, 14 Mar 2013 12:47:28 -0500
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.8; rv:17.0) Gecko/20130307 Thunderbird/17.0.4
MIME-Version: 1.0
To: internals@lists.php.net
References: <CADegSEbVU9MxX8e71tvCRLKN4SMLfcX+m=enKo13Wn=05QTEjA@mail.gmail.com> <op.wtxplogyc9evq2@nikita2206-debian> <CADegSEbK9=+OqPKqEM_-XyBk3K2cXUS_5+xiqXQnZq7foH2aaw@mail.gmail.com> <CAMyk_AZ_JDE9AZGqkqTxmFKe40CB9JU1iwzhTbmPMZoZnVgR_w@mail.gmail.com> <BLU0-SMTP195F04B804FBE9CC95227FCE2EC0@phx.gbl> <51420567.8000309@lerdorf.com> <BLU0-SMTP94C1C9962C7280A643F733E2EC0@phx.gbl>
In-Reply-To: <BLU0-SMTP94C1C9962C7280A643F733E2EC0@phx.gbl>
Content-Type: text/plain; charset=windows-1252; format=flowed
Content-Transfer-Encoding: 7bit
Subject: Re: [PHP-DEV] feature request : easy shared memory
From: larry@garfieldtech.com (Larry Garfield)

On 3/14/13 12:28 PM, Bob Weinand wrote:
> Am 14.3.2013 um 18:14 schrieb Rasmus Lerdorf <rasmus@lerdorf.com>:
>
>> On 03/14/2013 09:13 AM, Bob Weinand wrote:
>>> And there is no possibility to store the zval as raw binary data like in memory (deep copy?)
>>> So that you only have to copy from ram? And replace the pointers to the place in the string?
>>> This must be possible I think. And should be faster.
>>>
>>> shmop has to be opened on every request and only supports strings.
>>> APC, memcache,... can only save under serialized form which is slow.
>>
>> APC doesn't serialize most types. Only actual objects need to be
>> serialized because it is the easiest way to fully save and restore
>> objects. eg. calling their __sleep()/__wakeup() magic methods, etc.
>> Arrays are not serialized.
>>
>> -Rasmus
>
> Thanks, ..., okay, didn't know that.
>
> But even now I am in favor of a new keyword as it will be easier to have a reference to the
> shared memory (written in and reread from memory when modified) than every time refetching it
> when the shared memory block may have changed in an other program (what could really reduce
> race-conditions implicitly as as a developer you may forget to refetch the variable from shared
> memory). Yes, refetching always is already possible with an userland getter/setter, but I don't think
> it's best practice to do so in PHP...
>
> Bob Weinand

Sharing active memory between processes goes against the "shared 
nothing" design of PHP.  The lack of the feature you're describing is 
itself a feature. :-)

If you had real shared memory, then you're now writing a multi-threaded 
app.  Even if you aren't using threads per se it's the same level of 
potential for spooky action at a distance.  If your problem space really 
requires that (and there certainly are those that do), Java or NodeJs 
will suit you better because those are built specifically for a 
persistent-server model, rather than PHP's shared-nothing design. 
However, in practice most PHP/web applications don't need that, because 
HTTP is a stateless request/response system.  Shared-nothing more 
closely models what the actual environment is doing, and can still be 
very performant as long as you don't do anything naive.

If you're doing something stateful like Web Sockets, then you can run 
PHP as a cli application that is its own persistent server rather than 
as an Apache add-on.  For that, look at Ratchet: http://socketo.me/

--Larry Garfield