[omniORB] performance comparison

[Sai-Lai Lo]

"Gary D. Duzan" <gdd0@gte.com> writes:

> 
> In Message <199804171304.OAA01586@santaka.cam-orl.co.uk> ,
>    Sai-Lai Lo <S.Lo@orl.co.uk> wrote:
> 
> =>Apparently, there is a throughput test published on this page comparing
> =>omniORB 2.5.0 with Visibroker 3.1 and Orbix 2.3c:
> =>
> =>  http://www.kav.cas.cz/~buble/corba/graf/through.html
> =>
> =>Nice to know we consistently come on top and beat the other two by a huge
> =>margin on type Any :-)
> 
>    Some interesting data there. In particular, I'm amazed that ULong
> and Octet are actually faster using Any than the native types in
> OmniORB. Any idea why this would be?

Surely Any cannot be faster than native types. I guess it is just a small
margin of error in the measurement.

One point to note is that for small arguments the time to marshal the
data is very insignificant compare to other overheads introduced by the
runtime and the OS. As a rough estimate, around 100-150 micoseconds is the
fix overhead of the omniORB2 runtime, about the same amount is the OS
overhead. We are taking a performance hit in using a few try-catch loops
along the call-chain. 

What the test hasn't shown is how the ORBs scale as the number of
concurrent clients increases. We have been very careful to minimise thread
contentions and thread switching within the runtime. I believe this will
pay off as the load on the server increases.

You may notice that the only category where omniORB2 is slower than
Visibroker and Orbix is in array of basic data types. This is because all
arrays in omniORB2 are marshalled element by element whereas other ORBs may
have optimised to marshal the whole array in bulk. I can't be bothered to
do that optimisation :-)

Sai-Lai


-- 
Dr. Sai-Lai Lo                          |       Research Scientist
                                        |
E-mail:         S.Lo@orl.co.uk          |       Olivetti & Oracle Research Lab
                                        |       24a Trumpington Street
Tel:            +44 223 343000          |       Cambridge CB2 1QA
Fax:            +44 223 313542          |       ENGLAND