Re: Variability in one-way latency and jitter

[Marc Siegel <mlsiegel ll mit edu>]

Hi Bowie,

On Thu, 2004-07-22 at 19:10, Bowie Owens wrote:
> Marc Siegel wrote:
> 
> >Sorry for the reposting there.
> >  
> >
> Have you sent this to the ORBit2 list? All the marshalling, GIOP, and so 
> on is done in ORBit2 (C code). The wrapper layer for orbitcpp does 
> introduce inefficiency (converting C++ types into C types, some memory 
> allocation, ...) but it should be fairly predictable (aside from maybe 
> the memory allocation/deallocation) and there is no difference in the 
> wrapper between asynchronous and synchoronous.

Yeah I sent it to the list Orbit-list gnome org as well.

> >>
> No, I haven't seen this before. But I don't use one-way much and I 
> haven't been doing serious profiling of the project I use ORBit2 for. If 
> you can supply profile data indicating inefficiencies in the orbitcpp 
> wrapper, I can look at fixing that. But I suspect that isn't what is 
> causing the problem.
I'm sure it's not the wrapper, and MICO and TAO performed similarly. I'm
talking about latencies and jitter that vary 3 or more orders of
magnitude, very wild variation in local asynchronous performance. Maybe
it's due to a common implementation strategy for buffering or locking on
asynchronous calls.

If you know a specific person who might be able to offer some insight
into this, please let me know or forward to them?

Thanks sincerely

-- 
Marc Siegel
MIT Lincoln Laboratory
Group 63, Room C-439
mlsiegel ll mit edu 
(781)981-5434
--