[ntp:questions] (Software) timeserver for windows being broadcast-able incl. keys

Per Hedeland per at hedeland.org
Fri Mar 16 21:28:18 UTC 2007

In article <45FAAA30.2080401 at cag.zko.hp.com> Tom Smith
<smith at cag.zko.hp.com> writes:
>Erik wrote:
>> On 15 mrt, 17:21, Tom Smith <s... at cag.zko.hp.com> wrote:
>>>> broadcast key 1
>>>> broadcast key 1
>>>> broadcast key 1
>>>> broadcast key 1
>>> That's the right idea, but the second one above already
>>> includes the first.

How do you figure that? If (according to latest report) there are two
networks a) and b), the address doesn't "include" in any way.
is the broadcast address for a), is the broadcast address
for b) - and hosts on network a) will not see as a
broadcast address - in fact they will just see it as a random address on
a remote network. A broadcast address is either or one
where the network part matches exactly and the host part is all-ones.

>> there is one network-card in the server that connects to the network
>> and has access to the network segments mentioned above
>> The IP-data of this server (PC) is (ipconfig-output):
>> IP-adres . . . . . . . . . . . . . . .:
>> Subnetmask . . . . . . . . . . . .:
>> Standaardgateway . . . . . . . .:
>> Does this answer your question?
>Yes. That answers the question. With that configuration,
>in order to reach any of the clients, the packets that the
>server sends will have to be routed and cannot (usually) be
>broadcast. The server can (usually) only broadcast to its
>own subnet(s).
>That said, there are, of course, exceptions. I believe
>that if the subnets are, in fact, all on the same VLAN,
>you may be able to send a broadcast addressed to a
>network wider than the subnet defined by the server's
>netmask to any other network on the VLAN. In that case,
>you could use the single broadcast address
>to reach all of your clients. It might work and it might not.

No way *that* could possibly work - it's not a broadcast address on any
of the relevant networks. *If* the subnets are actually on the same
wire, what *might* work is to give the server an address in each network
(using "virtual interfaces" or "aliases" depending on the OS flavor),
and then specify all the individual broadcast addresses on their own
"broadcast" line in ntp.conf.

>The second exception is if your routers are configured
>route broadcast messages to be beyond the subnet
>on which they originate. In that case, you could again
>use the single broadcast address or
>the 3 individual broadcast addresses (2 through 4 in
>your list). Again, it might or might not work in your
>existing network configuration.

No - "directed broadcast" could work if the routers are so configured
(which they typically aren't, but that could perhaps be changed) - but
it still requires that ntpd actually sends out packets with each of the
per-network broadcast addresses. But here another problem arises - none
of the client-network broadcast addresses are broadcast addresses on the
server network. I'm not sure what ntpd will do with "broadcast" lines
with an address that isn't actually a broadcast address on any directly-
connected network - it would "just" need to send the packets out with
the given address and let the kernel routing deal with it, but in case
it searches for a local interface with the given broadcast address, it
won't find one.

>What will work, without question, is not using
>broadcast in the first place. You will have to work
>with the company who supplied your systems to fix
>the problem. You should continue this discussion
>with them. This is really no longer about NTP.
>It is about your network design.

Agreed - sorry for pushing the subject further off-topic.:-)

--Per Hedeland
per at hedeland.org

More information about the questions mailing list