"DBus Embedded" - a clean break

[Michael Meeks]

On Thu, 2011-01-20 at 12:17 -0500, David Zeuthen wrote:
> On Thu, Jan 20, 2011 at 3:09 AM, Rob Taylor <rob.taylor at codethink.co.uk> wrote:
> > Has anyone done benchmarks on GDBus vs libdbus serialisation performance?
> 
> Yes, GDBus in its currently unoptimized form is about 4x slower than
> libdbus (I think - the way this is measured is very sketchy). See here
> for details

	Right; I think this argument of "if you want performance, go cook your
own IPC mechanism" is not a very compelling one. I think a -much- better
argument is: "if you want performance, please help fix the existing IPC
mechanism".

	I see no real reason at all why the existing point-to-point dbus
implementation (particularly in a top-to-bottom solution like GDBus)
cannot be around 8x faster than GDBus is; and rather comparable to
dumping data directly to a pure socket with some horrendous hand-cooked
whatever.

	As such, unless you have some very simple, flat, nearly
uni-directional, trivial use case: "dumping megabytes of image data
between processes" - advocating direct socket use seems like rather a
retrograde step - surely ? :-)

	ATB,

		Michael.

-- 
 michael.meeks at novell.com  <><, Pseudo Engineer, itinerant idiot