Some comments on the D-Bus specification

Simon McVittie simon.mcvittie at collabora.co.uk
Thu Apr 23 04:38:31 PDT 2015


On 23/04/15 01:36, Thiago Macieira wrote:
> On Wednesday 22 April 2015 18:16:45 Simon McVittie wrote:
>> (This means that the dbus-daemon as currently implemented will stop
>> working after between 2**63 and 2**64 connections
> 
> The daemon could construct this from other components. For example, it could 
> take the monotonic clock's time of the connection and append the peer's PID, 
> which would be reasonably "forever" secure, even with reuse of the PIDs.

Sure, that would be one reasonable solution, if this needs fixing.

However, what we have at the moment can handle a new connection per
nanosecond for approximately 500 years of uptime, if my back-of-envelope
calculations are correct. I think that's already massively
over-engineered for our current and near-future requirements. The
reference implementation of D-Bus has many design and implementation
points that, with hindsight, were not ideal, but I think this particular
part holds up fine.

When we get closer to having fast enough computers (and IPC
implementations!) that we can initiate several connections per
nanosecond, or reliable enough computers that sustaining that sort of
rate over multiple decades is plausible, *that* is the time to be
replacing the current implementation of unique names :-)

-- 
Simon McVittie
Collabora Ltd. <http://www.collabora.com/>



More information about the dbus mailing list