Re: Isolated Network Drift Problem

Thank you for your detailed examples Steve. You have been very helpful.
I'm preparing to leave work for the Holiday weekend. If you are also a
US citizen, I'll wish you an enjoyable Thanksgiving Day.

../Cal

On Wed, 2008-11-26 at 18:06 +0000, Steve Kostecke wrote:
On 2008-11-25, Cal Webster <cwebster@xxxxxxxxx> wrote:

On Tue, 2008-11-25 at 21:10 +0000, Steve Kostecke wrote: [...]

Unfortunately, connection to the Internet is not an option I'm
allowed to consider.

Another option you could consider is the dumbclock driver. This uses
a serial link to transfer time stamps from one system to another.

Thanks but I'm afraid this would be of little use to me since most of
the systems are separated by brick and mortar, connected to each other
through DSL links on site.

This was a suggestion of a means of getting timestamps to the primary
time server in your time island. Distribution of the time to other ntpds
would occur over the network.

So, if I wanted to setup my 4 NTP servers to look to the one with the
best local clock as the master (if you've been following this thread)
but fall back to Orphan mode when it became unreachable, could you
give me an example of how the configs might look? I could later point
them to the one with the GPS time reference as master when I get it
setup.

My main concern is to continue to have an NTP server available in
each of 3 buildings even if they loose connectivity with each other
for some reason. I currently have at least one NTP service running in
each building, serving clients on MS Windows, Linux, and various Unix
flavors.

You need to set up the time servers for the orphan group in a "mesh"
(i.e. they all need to poll each other for the time). You may use any
NTP association mode for this purpose.

The following examples are incomplete; you need to add the drift file,
any desired restrictions, etc.

*** Unicast:

# Master Time Server *
# setting minpoll 4 makes this ntpd available to clients
# approximately 50 seconds after startup
server 127.127.1.0 minpoll 4
fudge 127.127.1.0

# Bldg A Time Server
tos orphan 10
server master.server.ip.address iburst
server bldg.B.ip.address iburst
server bldg.C.ip.address iburst

# Bldg B Time Server
tos orphan 10
server master.server.ip.address iburst
server bldg.A.ip.address iburst
server bldg.C.ip.address iburst

# Bldg C Time Server
tos orphan 10
server master.server.ip.address iburst
server bldg.A.ip.address iburst
server bldg.B.ip.address iburst

If you have additional time servers create similar ntp.conf files and
make sure that each servers knows about the others.

*** Manycast:

This mode provides automatic server discovery and creates unicast
associations. To use Manycast your server must be configured for to
support Multicast addresses _or_ you need a site-wide broadcast address
(i.e. every one on the same subnet). You also need to be comfortable
with setting up symmetric keys.

One advantage to using Manycast mode is that you can drop in additional
time servers without reconfiguring all of your existing time servers.

# Master Time Server *
# setting minpoll 4 makes this ntpd available to clients
# approximately 50 seconds after startup
server 127.127.1.0 minpoll 4
fudge 127.127.1.0
manycastserver broadcast.or.multicast.address

# Bldg Time Servers
# each "building time server" uses this file
tos orphan 10
manycastclient broadcast.or.multicast.address key N
manycastserver broadcast.or.multicast.address

* Note:

I am suggesting the use of the Undisciplined Local Clock on your Master
Time Server primarily beacuse The Undisciplined Local Clock has some
"knobs" which may be used to interactively trim its performance through
ntpdc. Orphan Mode, on the other hand, does not provide these knobs.

There is also a rootdispersion issue with Orphan Mode for stable
releases up to 4.2.4p5 and for development releases before 4.2.5p101. I
am currently conducting some tests.

.

Relevant Pages

  • Re: SYSVOL replication stops after DCPROMO
    ... unable to get the time from our central server. ... As the not finished SYSVOL replication is preventing the machine from ... Locator Flags: 0xe00001fc ... Preferred Time Server Name: \\AI2C001.auto.contiwan.com ...
    (microsoft.public.windows.server.active_directory)
  • Re: Primary WINS Server
    ... that Time Server option is not used in DHCP. ... The SNTP value on the server is as you expected, ... of the Linux box as the time server. ... is synching with the SBS. ...
    (microsoft.public.windows.server.sbs)
  • Re: Primary WINS Server
    ... The SNTP value on the server is as you expected, ... of the Linux box (which is not running a time server) as the time server. ... is synching with the SBS. ...
    (microsoft.public.windows.server.sbs)
  • Re: Ntp in peer configuration
    ... An "Orphan mesh" is the preferred way to handle your situation; ... server A iburst ...
    (comp.protocols.time.ntp)
  • Orphan Mode Configuration - unicast
    ... Here's a working orphan mode configuration using unicast mode for three ... server 192.168.0.2 iburst ...
    (comp.protocols.time.ntp)