[omniORB] Performance

[janssen.h at atlas.de]

Hi,
I'm still unsatisfied with the transfer rate of omniORB. Therefore I
have used a profiler to look behind the scenes and I find out that the
function timedwait (in file nt.cc) costs 50% of the runtime. Another 40%
are spent on the functions marshalChar and unmarshalChar (cdrstream.h).

My question is now, if there are maybe optimizations for the different
operating systems?

Thank you
 Heiko

-----Original Message-----
From: omniorb-list-bounces at omniorb-support.com
[mailto:omniorb-list-bounces at omniorb-support.com] On Behalf Of
janssen.h at atlas.de
Sent: Monday, February 06, 2006 10:04 AM
To: omniorb-list at omniorb-support.com
Subject: RE: [omniORB] Performance

Thanks for your help 

Heiko

-----Original Message-----
From: omniorb-list-bounces at omniorb-support.com
[mailto:omniorb-list-bounces at omniorb-support.com] On Behalf Of OKeeffe,
Michael K
Sent: Thursday, February 02, 2006 6:45 PM
To: omniorb-list at omniorb-support.com
Subject: RE: [omniORB] Performance


>
>Also check your IDL.
>For complicated data structures marshaling time is biggest
>consideration.
>If you use something like vector of anys with structs in it...
>

And related to that, the performance depends on the message sizes, see:
http://www.atl.external.lmco.com/projects/QoS/compare/dist_oo_compare_ip
c.html

Set the filter for two_hosts.

So it might be more useful to use an established benchmark.  More
reading:

http://nenya.ms.mff.cuni.cz/publications/TumaBuble-OverviewOfTheCORBAPer
formance.pdf

>
> 
>
>-----Original Message-----
>From: omniorb-list-bounces at omniorb-support.com
>[mailto:omniorb-list-bounces at omniorb-support.com] On Behalf Of Wernke
>zur Borg
>Sent: Thursday, February 02, 2006 5:21 AM
>To: omniorb-list at omniorb-support.com
>Subject: RE: [omniORB] Performance
>
> 
>In addition to what Alexander wrote, you could also think about oneway
>operations, which avoids the responses and thus shortens the latency
>considerably. It depends on the type of your application, however,
>whether or not that would be suitable.
>
>Wernke
>
>> -----Original Message-----
>> From: omniorb-list-bounces at omniorb-support.com
>> [mailto:omniorb-list-bounces at omniorb-support.com] On Behalf Of 
>> Alexander Haarer
>> Sent: 02 February 2006 11:08
>> To: omniorb-list at omniorb-support.com
>> Subject: Re: [omniORB] Performance
>> 
>> The throughput depends on two factors:
>> - network bandwith
>> - latency for one corba request/reply pair
>> 
>> Since the constant time that is spent with one request is 
>significant,
>
>> you should sent as much data as possible in one corba request to 
>> maximize the throughput.
>> 
>> janssen.h at atlas.de schrieb:
>> >
>> > Hi,
>> >
>> > has anybody experience with the transfer rate using 
>omniORB? I have 
>> > written a small program, which transmits different blocks of data 
>> > between two computers. The transfer rates are the following:
>> >
>> >  
>> >
>> > 100 MBit network = 8 MByte/s
>> >
>> > 1GBit network = 14 MByte/s
>> >
>> >  
>> >
>> > Why are the transfer rates so low?
>> >
>> >  
>> >
>> > Thank you,
>> >
>> > Heiko
>> >
>> > 
>> --------------------------------------------------------------
>> ----------

_______________________________________________
omniORB-list mailing list
omniORB-list at omniorb-support.com
http://www.omniorb-support.com/mailman/listinfo/omniorb-list


_______________________________________________
omniORB-list mailing list
omniORB-list at omniorb-support.com
http://www.omniorb-support.com/mailman/listinfo/omniorb-list