[omniORB] [giop 1.2 interoperability] Is omniORB compatible with GIOP 1.2 ?!?

[Duncan Grisby]

On Tuesday 16 September, Tomasz Bech wrote:

> More I work with omniORB more I like it and more hate Visisbroker. 

:-)

> > With short messages, less
> >than 8K, the whole message fits in a single network buffer, so there's
> >no need to fragment.
> >  
> 
> I found this problem much earlier, so I increased the buffer (*128) and 
> worked fine.
> (I had problem with marshaling long sequences).
> I don't know what's better - increase the buffer or switch back to giop 
> 1.0 - ??

I'd suggest switching back to GIOP 1.0. Increasing the buffer size
means omniORB uses a huge amount of memory for buffers, which is
probably worse than having it pre-calculate message sizes for GIOP
1.0.

> So I suspect also problem in omniORB - even that the buffer is huge
> CORBA::ULong giopStream::bufferSize = 8192*128;
> the data are still in this case streamed as several messages (my message 
> fits this 8192*128 bytes).
> The difference here is that long wstring is marshalled. Maybe there is 
> another constant which tells the maximum size of  wstring/octect<w_char> 
> or whatever.

The other parameter you're hitting is the directSendCutOff, which
determines when to start sending messages with direct sends from
application buffers, rather than going through omniORB buffers.

Cheers,

Duncan.

-- 
 -- Duncan Grisby         --
  -- duncan at grisby.org     --
   -- http://www.grisby.org --