[ejabberd] More virt host woes.

Tom Samplonius tom at samplonius.org
Mon Feb 11 07:26:17 MSK 2008

----- "Brian Cully" <bcully at gmail.com> wrote:
> 	Honestly, I have to think that the ODBC stuff is done this way on  
> purpose to drum up sales for Process One's enterprise server. Anyone 
> with even the smallest amount of experience knows that you keep a  
> single persistent pool of connections around for these kinds of  
> workloads.

  I have to strongly disagree with that.  I have experience with enterprise apps, and I know that pooling resources is not a "one size fits all" proposition.  There are a lot of reasons why you may want to have separate resource pools for separate domains.  The obvious being, is that each domain is on a different database server.  The other is resource isolation (ie. bulkheads).

  It is more reasonable to say, that ejabberd was designed for small numbers of domains (10 or so), each domain potentially very big.  In which case, the present architecture is a good fit.  You get to explicitly size the resource pool per domain, you can put each domain onto different db servers, and you get bulkheads between domains.

  I recommend "Release It!", by Michael T. Nygard, as a good book at enterprise deployments, and discussions about resource pools.  http://www.pragprog.com/titles/mnee



More information about the ejabberd mailing list