[omniORB] omniORB 2.6.1 + IRIX 6.4 (additional fixes)

Doug Anderson doug@clark.net
Thu, 10 Dec 1998 03:03:33 -0500


I was able to get omniORB 2.6.1 to work on IRIX 6.4
using the existing porting advice, but I needed to fix something
to get it to build 100%.

Underneath the following line in the IRIX platform file:

OMNIORB2_LIB += $(patsubst %,$(LibSearchPattern),omniORB2)

I added this _additional_ line below it:

OMNIORB2_LIB += $(patsubst %,$(LibSearchPattern),omnithread) -lpthread

This solves the problems in building (and running) the examples and also
the noted "nameclt" problem in the SGI README file.

Oh, another thing - I needed to grab the "mkdirhier" script off my DEC system
to get things going.  There wasn't one in my path, which was pretty standard,
so if someone knows where it is supposed to be on IRIX 6.4, let me know.
Otherwise, it might be a good idea to package up this script along with the 
distribution.

----  Now on a slightly unrelated subject:

By the way, omniORB is certainly one EFFICIENT ORB!!!  I am quite
impressed by the performance I've seen.  Especially on creating lot's of
objects.  It's ~10 times faster than TAO (single POA) on creating 10,000 objects,
and ~5 times faster than MICO on the same test, a simple factory interface that
returns a new object per create method call, all called in succession by a
single client thread.  It also beats Orbix, Visibroker, and ORBacus on the
same test, but I can't recall the exact stats except that of these three,
the closest one was half as fast as omniORB!!  :)  (on my DEC, I was able
to create ~1600 simple objects/sec on same machine with omniORB)

I understand why TAO probably doesn't accel on this type of test, being that
in a real-time system you usually can't afford to allocate lot's of new objects,
but I guess I wasn't expecting to see this much of a difference.  The difference
is even more dramatic if I increase the number of objects to 100,000 - try
50 times faster than the slowest.  I realize this is all subjective without
complete details, but still interesting.

Cheers,
Doug Anderson