[Solar-talk] Adapter-aware session package

[Paul M Jones]

On 11 Nov 2007, at 09:55, Richard Thomas wrote:

> User space session handlers tend to be slower for a couple reasons
>
> 1. Most people don't take into account session data doesn't change
> every page load, 99% of all user land session handlers I have seen
> read and write every single page load
>
> 2. If you don't store your data serialized you have to unserialize the
> data input and serialize the data output to conform with the built in
> session handler which adds extra overhead.
>
> 2 can't really be resolved without rewriting the whole ext/session,
> but 1 can be dealt with by hashing the data on read and only writing
> when the data has changed.
>
> You can also take that one step further, When you read check memcache
> first, if not in memcache then pull from file/db.. When you do need to
> write reset memcache.
>
> Using the above methods you can get your session data to the point
> where its always reading from memcache, only writing when data
> changes.

At which point it seems, you might as well be only working with  
memcache, and no file or DB intercessory layer.  The hashing you  
suggest is equally applicable to a memcache-only technique.


-- pmj