dbus performance information

Alban Crequy alban.crequy at collabora.co.uk
Thu Mar 20 10:30:59 PDT 2014

On Thu, 20 Mar 2014 17:24:22 +0100
VITIELLO Fabien <Fabien.VITIELLO at esg-group.fr> wrote:

> Hi,
> Several old benchmark exist over internet. For example, old
> comparison between CORBA implementation and dbus says dbus is 18
> times slower than CORBA. I'm looking for information about dbus
> performance. Do you have more recent information on this topic ? I
> didn't find answers in the documentation site except in the dbus FAQ
> question 7. It says dbus is about 2.5 slower than classic IPC
> mechanism. Is it an up-to-date information ?

> Moreover, I found an interesting topic concerning "FD passing"
> mechanism. Do you know if it is a way to speed dbus inter-process or
> if it's already implemented ? If it is better I was wondering how to
> activate this functionnality and how it works.

FD passing in D-Bus is already implemented. It is the feature to
transfer a file descriptor from one process to another by attaching it
to a D-Bus message. Internally, D-Bus uses the ability of Unix sockets
to transfer a file descriptor in sendmsg(2)/recvmsg(2)'s ancillary data.

This feature is used by Tracker for performance reasons: it creates a
pipe and sends one of the pipe's file descriptor through D-Bus (with
GDBus) and then use the pipe to transfer data rather than D-Bus.

It could be to transfer a file descriptor of a specialized type. Bluez
uses it to give the AF_BLUETOOTH socket file descriptor to oFono (with

If you use GDBus, you can check the documentation of *_fd* functions:

Best regards,

More information about the dbus mailing list