[ntp:hackers] Serial port jitter
David L. Mills
mills at udel.edu
Thu Jul 6 23:32:57 UTC 2006
Some motherboards make better timekeepers than others.
Sun Ultra/5_10 pogo and Intel P2 rackety are in the low microseconds
with the serial port and PPS port with no special treatment, like
disabling the hardware FIFO. The pogo cat nees fudge time1 .03154 s to
shim the serial port to match the PPS, while the rackety cat needs only
.009445 s. Something is eating some twenty millisconds in pogo, which
seemsa rather strange, but both show very low jitter on either the
serial or PPS ports.
On pogo the PPS signal is connected via dedicated serial port, while on
rackety it is connected by a parallel port. Each is connected to a
different GPS receiver with claimed accuracy of 100 microseconds.
Offsets relative to both of our TrueTime GPS-in-a-box is a few tens of
microseconds. These are separted by a couple miles of Etherstuff bushels
of hubs, routers and fiber. What is surprising is the same few tens of
microseconds, both in relative offset and jitter, are displayed for
three of the NIST servers, one in Boulder, another at Microsoft and the
third in Alaska (!). Not too shabby.
Sun Blade 1500 deacon connected to an Arbiter GPS receiver with PPS,
both connected by dedicated serial ports. It behaves much like pogo and
rackety. Not so 2.8 GHz Dell 370 P4 mort connected to a WWVB receiver
with PPS much like rackety. The serial port jitter is huge compared to
the other machines, 10-20 ms. If this is due to the FIFO, the FIFO would
have to have 20 stages, more than those UARTs I know.
I searched the FreeBSD documentation looking for the little magic that
disables the FIFOs, but couldn't find it anywhere. Does someone here
know the magic potion?
More information about the hackers