Hi Felix:
 
I don't have any hard data on the impact of the performance improvements
in TIPC 1.7.5.  I took a quick look at the change history for TIPC 1.7,
and can see several changes that could conceivably relate to the sort of
traffic you're using, but it's really tough to predict the outcome you
will see.  My own experience in testing the performance of message
passing software (be it TIPC, IP, or something else) is that the results
you obtain can vary significantly, depending on the exact nature of the
testing being performed (where variables include CPU architecture,
processor speed, interface speed, message size, burstiness of traffic,
bidirectional vs. unidirectional traffic, and so on).
 
That being said, I always encourage users to upgrade to the latest
version of TIPC if they can, so that they can take advantage of the
various improvements that have been made to the product.  In addition
the benefits provided from the bug fixes that were made, some of the
enhancements that have been made (such as the ability to have separate
threads send and receive on the same socket) may potentially open the
doorway to improved performance.
 
Regards,
Al

________________________________

From: Nayman Felix-QA5535 [mailto:[EMAIL PROTECTED] 
Sent: Wednesday, February 06, 2008 12:24 PM
To: Stephens, Allan; [email protected]
Subject: RE: [tipc-discussion] Performance improvements of using TIPC
1.7.Xversus TIPC 1.5.12


Al and all,
 
Well since our application cannot run in the kernel, I guess we'll need
to concentrate on determing whether it makes sense to upgrade to TIPC
1.7.5.  The release notes seem to mention performance improvements.
Most of our messages are around 2K in size (and as I mentioned before
sent and received in connectionless mode).  
 
Would we expect to see a measurable improvement in performance? 
 
 
Thanks for any input,
Felix

________________________________

From: Stephens, Allan [mailto:[EMAIL PROTECTED] 
Sent: Wednesday, February 06, 2008 11:01 AM
To: Nayman Felix-QA5535; [email protected]
Subject: RE: [tipc-discussion] Performance improvements of using TIPC
1.7.Xversus TIPC 1.5.12


Hi Felix:
 
TIPC's native API is definitely capable of providing better performance
than the socket API.  However, it is only available to application code
running in the kernel, so it may not be a trivial matter to convert a
socket based application running in user space to take advantage of it.
The performance gain comes from two major differences from the socket
API: 1) the elimination of the overhead inherent in the socket API
(which introduces a number of nested function calls, as well as a system
call, into the transmit and receive paths), and 2) the ability to
process incoming messages in the context of the TIPC tasklet (rather
than in a user-space process).
 
A large chunk of the native API has undergone thorough testing due to
the fact that it is used by TIPC's socket API.  Some of the remaining
portion is used within TIPC itself, and has been tested in that manner.
The final portion has received only limited testing because few TIPC
users seem to use the native API, preferring to write their applications
in the more familiar confines of user space.
 
Regards,
Al

________________________________

From: [EMAIL PROTECTED]
[mailto:[EMAIL PROTECTED] On Behalf Of
Nayman Felix-QA5535
Sent: Wednesday, February 06, 2008 10:56 AM
To: [email protected]
Subject: [tipc-discussion] Performance improvements of using TIPC
1.7.Xversus TIPC 1.5.12


Hi,
 
We're trying to see if we should upgrade to TIPC 1.7.X from TIPC 1.5.12
for performance reasons (and we'll do some benchmark testing to see what
the difference is).  
But I'm wondering what other people's experience has been.  Have they
seen a marked improvement?   We're running connectionless traffic (the
socket is blocking) on a Linux box. 
 
Also, we're currently using the socket API.  I remember reading
somewhere that the native API would provide for better performance.  Is
that true?
I was also under the impression that the native API was still under
development and not as thoroughly tested/used as the socket API.  Are
there issues/difficulties with using the native API on either 1.5.12 or
when upgrading to 1.7.X?
 
 
Thanks for any insights,
Felix
-------------------------------------------------------------------------
This SF.net email is sponsored by: Microsoft
Defy all challenges. Microsoft(R) Visual Studio 2008.
http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
_______________________________________________
tipc-discussion mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/tipc-discussion

Reply via email to