[Accessibility] [Fwd: [g-a-devel] Selector interface and CORBA performance debate]

[Peter Korn]

Greetings,

In various discussions about implementing GNOME accessibility, one =

concern that has been commonly and repeatedly raised is about the =

performance of CORBA for IPC.  DBUS has been commonly suggested as a =

higher performance mechanism to use.

Frank Duignan completed an analysis of CORBA performance as compared to =

DBUS and DCOP (the KDE/Qt IPC mechanism).  His results are interesting.

His summary conclusions:
  DCOP is approximately 3 times slower than CORBA
  DBUS is approximately 18 times slower than CORBA

This is based on the sample calls he made to void, int, and string =

functions across process boundaries on the same CPU.  See his e-mail =

message attached, and his report at =

http://eleceng.dit.ie/frank/rpc/CORBAGnomeDBUSPerformanceAnalysis.pdf =

for more details.  Clearly his tests are somewhat simple - they don't =

mimic the more complex interactions and "lengthier conversations" that =

we have in AT-SPI; further complicating factors mean these performance =

numbers likely won't translate directly to an AT-SPI implementation in =

DBUS or DCOP.  And there are any number of places we can look to =

optimize IPC performance besides moving from CORBA to DBUS/DCOP/whatever.

Nonetheless, it does suggest that CORBA itself isn't obviously a bad =

choice from a performance perspective...



Regards,

Peter Korn
Accessibility Architect,
Sun Microsystems, Inc.
-------------- next part --------------
An embedded message was scrubbed...
From: Frank Duignan <frank.duignan at dit.ie>
Subject: [g-a-devel] Selector interface and CORBA performance debate
Date: Fri, 26 May 2006 09:17:38 +0100
Size: 6864
Url: http://lists.linux-foundation.org/pipermail/accessibility/attachments/=
20060526/5290f66f/attachment.eml