Hi,

On Feb 12, 2013, at 10:13 AM, Sergey Dobrov <bin...@jrudevels.org> wrote:

> On 02/12/2013 02:34 PM, Simon Tennant wrote:
>> I understand that you want to do lookup in PEP. I'd be very wary of
>> depending on it to build a full scaled social platform.
>> 
>> We tried using PEP to build buddycloud. We created a bunch of nodes for
>> each and would push out updates based on presence. This kinda worked
>> with the following caveats::
>> - it was server specific (any changes to PEP had to be made to each
>> server)/you loose the portability of XMPP components
> If the features will be described in XEPs any server will be able to
> implement them. But they are needed to be interested in it. The interest
> may be cause with interesting and powerful user applicable protocols
> like XEP-277. But we are really need to make XEP-60 and PEP more usable
> for this kind of things.
> 
> Also, I already told that we need a standard that will allow services
> like PEP to be external like XEP-114 allows gateways to be external. I
> still thing this is the most priority task for pubsub development.
> 
>> - scaling and performance (this could have been an issue with ejabberd
>> at the time).
> Yeah, this is an ejabberd issue they are hoping to fix in 3.0.
> Unfortunately, we are waiting for 3.0 too long.

yes scaling and performance issues in 2.x is quite known… And I don't even 
think that Ejabberd 3.0 will be open sourced. You should take a look at 
mongooseIM, that is an erlang solutions open source fork.

> 
>> - difficult to build any extra business logic into PEP. For
>> example:"this set of users can reply to my posts"
> I don't see the problem here technically. This is impossible with
> current standards but technically it's not a problem at all.
> 
>> 
>> I think that the OneSocialWeb guys ran into similar issues with their
>> PEP-based service.
>> 
>> Our XMPP architecture for buddycloud has shifted: build as much as
>> possible outside of the XMPP server core. We use components to scale out
>> services and use the XMPP server just as a router (the famous "xmpp
>> middleware" topic). The benefit of using components for your service is
>> that running Movim will have the ability to switch out their XMPP server
>> as their needs change.
>> 
>> One solution that might give you some milage: using DISCO to point to a
>> pub sub component that is nominated for social activities? (Something
>> like https://buddycloud.org/wiki/XMPP_XEP#buddycloud_Server_Discovery).
>> This will give you much more flexibility in the future and nto tie your
>> solution to a particular XMPP server vendor.
>> 
>> S.
> 
> 
> -- 
> With best regards,
> Sergey Dobrov,
> XMPP Developer and JRuDevels.org founder.

Reply via email to