Gary E. Miller gem at rellim.com
Wed Jan 24 18:57:47 UTC 2018

Yo All!

Sorry for the corss post.  I mentioned this on the gpsd list and thought
some folks here might also be interested:

I just pushed clock_test.c to git head in contrib/.  Basically, the
test runs two back to back clock_getttime(), then waits a bit for the
CPU cache to clear.  The tests below are for 10,000 samples, with 10
milli sec between tests.  Needless to say the data is a bit jumpy, so
I had to show the standard deviation.

Here is one set of VERY preliminary results:

Intel CPU:

# ./a.out -n 10000
samples 10001, delay 10000000 ns
min 13 ns, max 13440 ns, mean 131 ns, median 24 ns, StdDev 964 ns

Arm 32 bit, 32 bit Gentoo, kernel like almost all RasPi distros:

# ./a.out -n 10000
samples 10001, delay 10000000
min 104 ns, max 22969 ns, mean 169 ns, median 156 ns, StdDev 296 ns

Arm 64 bit, 64 bit Gentoo.  Dev kernel.

# ./clock_test -n 10000
samples 10001, delay 10000000
min 156 ns, max 6667 ns, mean 178 ns, median 157 ns, StdDev 113 ns

Things to note:

No wonder people have such a problem breaking the 1 micro second
barrier on PPS!

Notice how much arm is better than amd64.  And how much better arm64 is
than 32 bit arm.

Gary E. Miller Rellim 109 NW Wilmington Ave., Suite E, Bend, OR 97703
	gem at rellim.com  Tel:+1 541 382 8588

	    Veritas liberabit vos. -- Quid est veritas?
    "If you can’t measure it, you can’t improve it." - Lord Kelvin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 488 bytes
Desc: OpenPGP digital signature
URL: <https://lists.ntpsec.org/pipermail/devel/attachments/20180124/78826266/attachment.bin>

More information about the devel mailing list