[omniORB] Socket reuse by omniORB?
Duncan Grisby
duncan at grisby.org
Thu Jan 24 14:29:43 GMT 2013
On Thu, 2013-01-24 at 05:07 -0800, Igor Polonsky wrote:
> 1) Having created a connection, Visibroker will hold to it
> indefinitely. But omniORB appears to drop any connections that have
> been idle for about a minute or so. Could you please confirm if that's
> really the case? And how can we fine-tune this behavior, if need be?
Yes, that is the case. See the documentation here:
http://omniorb.sourceforge.net/omni41/omniORB/omniORB008.html#htoc100
The configuration parameter that's important is outConScanPeriod, which
defaults to 120 seconds. If you set it to zero, idle connections will
never be closed.
> 2) When omniORB creates a new connection to a certain CORBA server, it
> seems to reuse any open connection to the same server - even if that
> connection has been created by an unrelated process, such as a
> Visibroker-based client running on the same machine. This brings a few
> questions:
>
> - Could you please confirm if this is really what's happening? Or am I
> misinterpreting the observed behaviors?
No. There is no way that omniORB could do such a thing. What are you
seeing that makes you think that is the case? Certainly the operating
system will reuse client-side port numbers over time, but only once the
previous connection from that port is completely closed down.
> 3) Can anyone please suggest some good reading on this subject? I'd
> like to understand how apps such as CORBA clients interact with the
> Windows OS when they create (and share?) connections. And also how
> CORBA connections consume other system resources, such as NP memory
> pool, etc. I'd like to become a slightly less dumb user of CORBA.
Most of these concerns are specific to a particular CORBA
implementation. There is quite a lot of detail about how omniORB behaves
in its manual, in particular chapter 8 that talks about connection and
thread management.
Cheers,
Duncan.
--
-- Duncan Grisby --
-- duncan at grisby.org --
-- http://www.grisby.org --
More information about the omniORB-list
mailing list