Re: [gpsd-dev] register keyword?

[Gary E. Miller]

Yo Kurt!

On Sun, 21 Jan 2018 18:43:20 -0800
Kurt Schwehr <address@hidden> wrote:

> Is there still a reason to keep the register keyword in gpsd?

Maybe.  It is in the critical path of some time measurements.

Does it make a measureable difference?  Hard to say.

Might be worth running some tests.

I've been running some timing tests on a different section of code.
Nothing I'm ready to commit to git yet, but already interesting
data.

I've been looking at the latency of clock_getttimme().  I have always
suspected, but never been able to prove, that the latency of that
system call on Intel 64 bit, was more than on arm32.  Well, I was
sorta right.

The mean latency on the one Intel I have tested is instead slightly less
than the arm32.  But the Standard Deviation is way worse.  So there
is now hard data why a RasPi can measure PPS better than an Intel.

In other news, I'm also playing with an arm64 distribution.  It is
indeed faster than arm32 for the same distro, and also lower latency
and lower standard deviation on clock_gettime().

So, back to your question, yes, these little tweaks matter, and need
to be tested before changing them.

RGDS
GARY
---------------------------------------------------------------------------
Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
        address@hidden  Tel:+1 541 382 8588

            Veritas liberabit vos. -- Quid est veritas?
    "If you can’t measure it, you can’t improve it." - Lord Kelvin

pgpcazPL2_TCB.pgp
Description: OpenPGP digital signature