🚀 go-pugleaf

RetroBBS NetNews Server

Inspired by RockSolid Light RIP Retro Guy

Thread View: comp.protocols.time.ntp
4 messages
4 total messages Started by system PRIVILEGED account Tue, 01 Mar 1994 16:22
NTP in demand dialup network
#1094
Author: system PRIVILEGED account
Date: Tue, 01 Mar 1994 16:22
56 lines
4036 bytes
Apologies to comp.protocols.time.ntp readers if this problem is either
(a) obvious or (b) boring.

Executive summary: how do I synchronise a network with the Internet
servers if I only have a demand-dial link, not a permanently connected one?

We run a small network of UNIX boxes, mainly running things like SunOs V4,
in which we use xntp V3.3 to keep a decent local idea of time for the
purposes of logging events and doing things like "make".

The structure of the network's ntp subnet is straightforward: every machine
is running xntpd.  One machine (yankee, a Sun-3) has "the clock" (its xntpd
synchronises with 127.127.1.10) and the rest synchronise to it.  This
met our first set of objectives (time-like behaviour of the local clocks)
fairly easily.  Of course, the machines don't know whether the time they hold
in common is correct in any sense, but they do end up closely synchronised.

Recently, we got a dial-up connection to the Internet, and it occurred to me
that it would be useful to make use of the net time servers to get our
internal network running on "real" time as well as just being internally
consistent.  Our IP supplier runs a stratum 2 service, so we don't even have
to go far afield to get "good enough" time for our purposes.  We are running
the dialup service off the same Sun-3 that our ntp subnet synchronises off,
so it seemed reasonable to try and do the synchronisation of our net by
running ntpdate each time we connect.  This would, I thought, slew the
local clock on yankee and the rest of the network would follow. (Obviously,
I didn't want to tell yankee's xntpd about the internet time servers because
it would end up prodding them rather often to find out what the time is,
and the implied local telephone calls are not free round here.  Not to mention
that there may be a 50-second asymmetric delay for the ntp packets because
of dialling the phone.)

Unfortunately, ntpdate won't set the clock on yankee if it notices that
the ntp port is in use (xntpd is running).  I can't stop the xntpd on
yankee or the rest of my network won't be able to synchronise with it.

Next, I tried running ntpdate in debug mode, which tells me what adjustment
it would like to make.  I tried to apply this adjustment using "date -a"
to slew the clock slowly round to the new value, but I get some "adjustment
didn't complete" messages from xntpd, presumably because, even though it is
taking its time from the local clock on yankee, it is also continuously
"correcting" that local time (by zero-size adjustments) to correct for the
(zero) drift it believes it has (!)

Next, I tried using "rdate" rather than "ntpdate" to get the time from the
network servers.  Although this works in the sense that yankee's time (and
therefore the network time) is now always within a couple of seconds of the
rest of the world, it is very jittery and the machines sometimes even run
backwards a half second or so.  Nasty.

I'm out of clues at this point.  Is what I am doing really such a misuse of
xntp that no-one has been through this before?  Any suggestions welcome
(except the ones about buying a permanent internet link to replace our
demand-dial one, and the ones about buying a radio clock...)

	-- Ian
Re: NTP in demand dialup network
#1095
Author: Alan Barrett
Date: Thu, 03 Mar 1994 08:17
23 lines
1874 bytes
In article <CLzutI.586@threel.co.uk>,
iay@threel.co.uk (system PRIVILEGED account) writes:
> Executive summary: how do I synchronise a network with the Internet
> servers if I only have a demand-dial link, not a permanently connected one?

I haven't run NTP ovr a demand-dial link, but I have run it over
a permanent link whose quality is usually abyssmal, but which has
occasional periods of decent quality.

If you configure a system with a local clock (127.127.1.something)
and with two or three low-stratum servers at the far side of the
dialup link, then I think you will find that xntp3 is able to synch
with the remote servers while the link is up, and free-run from the
local clock when the link is down.  I guess that periods of up time
shorter than about 15 minutes would probably not be enough to establish
synchronisation, but longer periods should be fine.

You would probably want to teach the dial-on-demand software that the
presence of NTP packets is not a sufficient reason to initiate a dialup.

--apb
Alan Barrett, Dept. of Electronic Eng., Univ. of Natal, Durban, South Africa
RFC822: barrett@ee.und.ac.za
Re: NTP in demand dialup network
#1096
Author: Alan Barrett
Date: Thu, 03 Mar 1994 08:42
13 lines
1372 bytes
In article <CLzutI.586@threel.co.uk>,
iay@threel.co.uk (system PRIVILEGED account) writes:
> Next, I tried running ntpdate in debug mode, which tells me what adjustment
> it would like to make.  I tried to apply this adjustment using "date -a"

xntpdc will let you write to the frequency component of the loop filter,
using the "fudge 127.127.1.whatever time1" command.  One could easily
hack the source to give xntpdc access to the offset component of the
loop filter, but I don't know whether that would be a good idea.

--apb
Alan Barrett, Dept. of Electronic Eng., Univ. of Natal, Durban, South Africa
RFC822: barrett@ee.und.ac.za
Re: NTP in demand dialup network
#1097
Author: Alexander Dupuy
Date: Thu, 03 Mar 1994 19:25
17 lines
1769 bytes
I posted a somewhat longish message a month or so ago about running
NTP over demand-dialup IP.  You can check the ppp-users archive at
ftp.morningstar.com, or the NTP mailing list archive (is there one?).

The brief summary is that you should use xntpdc, not ntpdate,
synchronized with a few secondaries (if you can teach your demand dial
software not to bring up the link just because there's an NTP packet
which wants to go out).  Set up a local clock "peer" at stratum 5 or
6, so that xntpdc will stay synchronized when the link is down.  Link
up times of only slightly more than 5 minutes appear to be sufficient
to keep my IPX within 50 msec of our secondaries, once the drift rate
is accurately calculated (this may take a day or two).

@alex
--
inet: dupuy@cs.columbia.edu
Member of the League for Programming Freedom -- write to lpf@uunet.uu.net
Thread Navigation

This is a paginated view of messages in the thread with full content displayed inline.

Messages are displayed in chronological order, with the original post highlighted in green.

Use pagination controls to navigate through all messages in large threads.

Back to All Threads