Hmm ! very very interesting. Well your implementation is what I
thought for one of my systems and didn't implement. Many of my designs
just die in my head though :-)

I guess it was even discussed in one of ML's related to persistent queues.

Wish I were an Entrepreneur, it would have been one of my product :-)

Its worthwhile capturing a sample implementation in our documentation.

Emmanuel, anything on this from ApacheDS side, like something that can
showcased

thanks
ashish

On Thu, Aug 27, 2009 at 11:20 AM, Christopher
Popp<[email protected]> wrote:
> Just thought I might share some of our experience with using MINA for 
> HA/scalability...
>
> What we actually do is have multiple servers running MINA that are simply 
> responsible for maintaining client TCP connections and processing received 
> operations.  We also have some servers that run front end applications that 
> are responsible for authentication and receiving operations.  All of this is 
> tied together using JVM level clustering through Terracotta.  Each connected 
> client in the system gets a queue, and the particular system that the client 
> is connected to registers a callback for new messages in that queue.  The 
> queues are optionally persistent for us, so that even if a client 
> disconnects, and reconnects to a new MINA system, the operation will be 
> processed once the new system registers interest in the client's still 
> existing queue.
>
> We've run this in EC2 before and it seems to work rather well.  It requires 
> some work on top of MINA to get the HA/Scalable aspect, but it wasn't too bad 
> to do.
>
> Chris Popp

Reply via email to