On Wed, Jul 01, 2015 at 09:15:20AM -0400, Christopher Hilton wrote:
On Jul 1, 2015, at 9:01 AM, Michael Stone <[email protected]> wrote:
On Wed, Jul 01, 2015 at 08:30:25AM -0400, Christopher Sean Hilton wrote:
The net4801 is supposed to be perfect as an inexpensive gps driven
clock for ntpd. But I thought that having CPU_ELAN and CPU_ELAN_PPS in
the kernel added precision to the timekeeping?
That's the net4501, not the net4801. FWIW, without replacing the clock and using a
custom ntpd the 4501 is ok as an NTP server but not spectacular. The limited memory
and cpu hurt if you run a current OS & ntpd. Remember, nobody's actively
targeted that hardware in a decade. The net4801 will probably do as well in the
real world even without the elan timer registers.
I assumed that a gps with PPS output was required.
I'm not entirely sure what you mean. Most hardware can handle PPS, over
the serial port if nothing else. It's typically implemented in ntpd,
possibly with gpsd providing abstraction. The net4501's timestamp
registers provide better theoretical performance than a serial port PPS.
In theory it's also possible to use the GPIO pins on a net4801 to
provide the PPS, but that still doesn't have anything to do with the
kernel PPS functionality tied to the elan timestamp regiesters. In the
real world NTP servers are generally used to serve time over the
network, and a 15 year old embedded processor is somewhat slow at doing
network things; the network jitter will far outweigh any theoretical
accuracy of the time source.
Mike Stone
_______________________________________________
Soekris-tech mailing list
[email protected]
http://lists.soekris.com/mailman/listinfo/soekris-tech
