Hello,

Is there on a git repo somewhere already ?
I'd be very interested in having a look at it.

Kr,

Diego Duclos


On Thu, Apr 10, 2014 at 9:53 AM, Pieter Hintjens <[email protected]> wrote:

> Hi Michael,
>
> This sounds great. My advice is to make small steps, solving one
> problem at a time, rather than aiming for ideal solutions. I'd love to
> see compile-time pluggable protocols come onto master, and be used,
> before we think about making this a runtime thing. Keep things simple,
> don't solve problems we don't have (yet).
>
> -Pieter
>
> On Thu, Apr 10, 2014 at 5:21 AM, Michael Holmwood <[email protected]>
> wrote:
> > Greetings All,
> >
> > I am currently involved in a project that aims to add SCTP as a transport
> > protocol to ØMQ. So far, we have managed to get SCTP working fine, we
> just
> > needed to add the ability to modify heart beat and add addresses for
> > multi-homing. The multi-streaming part is going to be the most difficult,
> > but I don't think we will be having any problems adding this.
> >
> > Back in February, I recall a message that was posted on the mailing list
> > regarding pluggable transport protocols. Having spent some time looking
> at
> > the ØMQ source code, I determined that I could add this without too much
> > trouble. So we have made some of the necessary changes, and have both
> SCTP
> > and TCP working as pluggable protocols, with TCP being the default
> option.
> > We have tested the performance with pluggable and non-pluggable (with
> TCP)
> > and noticed no difference.
> >
> > At the moment, there is no way to add protocols after ØMQ is compiled,
> which
> > is the next thing we needed to look at. We have considered a few options:
> >
> > 1) Leave it the way it is, having protocols only added when ØMQ is
> compiled.
> > I have seen something similar used in the UDT library, but this was with
> > pluggable congestion control algorithms. http://udt.sourceforge.net/
> >
> > 2) Provide something in the API to enable protocols to be added using
> shared
> > libraries.
> >
> > 3) Provide a mechanism to register new protocols via the API using
> factory
> > objects/functions.
> >
> > I am quite keen on option 3, however it would be good if I could get some
> > input from the community before proceeding. One thing to note is that it
> > only currently works with Linux, as this is what our client has
> requested,
> > but would look at modifying the code to work with both.
> >
> > The interface that I have created requires that the protocols be added
> using
> > classes created in C++, so I guess I might have to look at modifying it
> to
> > work with C or provide a mechanism to provide the appropriate functions
> > using pointers.
> >
> > Regards,
> >
> > Michael Holmwood.
> >
> >
> >
> > _______________________________________________
> > zeromq-dev mailing list
> > [email protected]
> > http://lists.zeromq.org/mailman/listinfo/zeromq-dev
> >
> _______________________________________________
> zeromq-dev mailing list
> [email protected]
> http://lists.zeromq.org/mailman/listinfo/zeromq-dev
>
_______________________________________________
zeromq-dev mailing list
[email protected]
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

Reply via email to