Posted by Jochem Maas on 10/21/86 11:36

Mark wrote:
> Jochem Maas wrote:
>

.....

>
> That's sort of the problem with names, it is hard to convey ideas. MCache
> (formally msession) has a history. I originally wrote the code as a way to

there is a php extension called 'mcache' with is a wrapper for memcache -
but this is not the same as your tool right? in which case I doubly recommend
another name :-)

> share data for a beowulf cluster. Later on I was working on a site that had

the words 'cluster' may very well be something to think about .. e.g. 'ClusterCache'

> multiple PHP web servers and realized it was the exact same problem domain.
> Multiple machines at any one instant mostly working on different data but
> requiring access to serialized shared data or a central store of data.
>
> This is a HUGE problem once you get past one PHP server. Even with one
> server you still need a database to manage your poll. You can't do this:
>
> $var = sql->exec("select poll_count from poles where pollid='foo' and
> name='bar'";
> $var++;
> $var = sql->exec("update poles set poll_count = $var where pollid='foo' and
> name='bar'";
>
> In an active site you will have concurrency issues and lose counts even on a
> single machine.
>
> So you'll have to do something like this:
>
> sql->exec("update polls set poll_count = poll_count+1 where pollid='foo' and
> name='bar'";
>
> $var = sql->exec("select poll_count from poles where pollid='foo' and
> name='bar'";
>
> That is *always* two database operations. Best not get slashdotted! With
> MCache, it works like this:
>
> $var = mcache_inc("foo','bar');

that is nice... :-)

>
> Now, depending on your strategy, there will may be no hits your database or
> (at most) one. How many times have you seen a site DOA because the database
> (typically MySQL) has crapped out?
>
> MCache on a modern computer, handling the session storage and retrieval, and
> a one or two variable calls like the above, can handle (depending on
> session size, network speed, etc.) upwards of 5000 operations a second. It
> is designed to be fast!!! Granted, if the session information is larger
> than one TCP/IP packet, it is slower, but still way faster than a database.
>

I just checked some session file sizes on a 'heavy' site I built. the largest
file was 332607 bytes; which very roughly speaking would require about 220 packets.

is such a large session file feasible with you tool? (regardless of whether
one might suggest that I have a design flaw in the site in question)

>
>
>
>>>
>>>>is this better than using the 'files' session handler and setting the
>>>>session save path to a tmpfs filesystem (in effect a RAM disk) in a
>>>>signle server env.?
>>>
>>>
>>>This is not really intended for a single server site, but could provide
>>>some benefit as a way of abstracting file or SQL session storage.
>>
>>ok; although that doesn't answer my question.
>>another thing if it's not really intended for a signle server site then
>>youre audience here is rather limited - I'm guessing that there are not
>>many here running distributed/load-balanced systems.
>>
>>I don't atm; but it almost makes me want to go out and get a server
>>to try it out.
>
>
> Well, using the poll operation detailed above, you can see relative
> advantages. It is faster than the PHP file based sessions. It can
> periodically save session data. It does its own garbage collection. It can
> be used to call external functions. It will eventually (there now, needs
> testing and debugging) even have a SQL based write through caching plug-in
> that will allow an amount of database abstraction, or even session storage
> abstraction.

I for one will definitely be watching your space :-)

>
>
>>>>>the word out and get some testers.
>>>>>
>>>>>The server itself is GPL, the PHP extension is under the PHP license. I
>>>>>will be checking it in to the PHP main CVS repository when I'm pretty
>>>>>sure that it will play nice.
>>>>>
>>>>>The server can be accessed via CVS from www.mohawksoft.org.
>>>>>
>>>>>If you have any questions please feel free to ask.
>>>>>
>>>
>>>
>

[Back to original message]