Help Broadcasting a UDP packet on the LAN:URGENT

[Charles Swiger]

On Thursday, October 23, 2003, at 03:43 PM, Barney Wolff wrote:
> On Thu, Oct 23, 2003 at 02:23:57PM -0400, Charles Swiger wrote:
>>> What are you going to do when IPv6 comes into more general use, since
>>> it has no broadcast address?
>>
>> Are you asking what a IPv4-to-IPv6 translator (like gif?) should do,  
>> or
>> are you worried about the case of a machine configured for IPv6 only
>> and not for IPv4?  I expect that most people will be using IPv4 for
>> quite some time; we don't have to do something for the IPv6-only case
>> to still have this be useful.
>
> My expectation is the same as yours, but I strongly believe that
> anyone doing a new design that deliberately ignores IPv6 is being very
> shortsighted.  "Quite some time" is now only years, not decades.

There are people using IPv6 now, certainly.  There are also a number of  
possible opinions with regard to the utility of IPv6-- whether IPv6 can  
solve problems sufficiently better than IPv4 (or IPv4 plus NAT,  
particularly) as to make IPv6-only networks a reasonable alternative to  
existing IPv4 networks is an open question.

Perhaps I am being shortsighted in my focus on IPv4, but I don't think  
that I am ignoring IPv6 so much as concluding that IPv4 all-ones  
broadcast doesn't need to have a precise cognomen in the realm of IPv6  
to still be useful.  No doubt the kind and friendly people on this list  
will continue to point out any major bits of stupidity or  
shortsightedness on my part in the future.  :-)

>> Multicast and broadcast addressing are working at layer-3, but the
>> point of using VLAN tags is to create logically 'seperate' networks
>> where the flow of traffic is being handled/segregated at layer-2  
>> rather
>> than layer-3.
>
> VLANs are meant to allow segregating a physical switch into multiple
> logical switches.

That's the primary usage, certainly.  You can use VLAN tags on a  
non-switched environment, however, just as you can send packets for  
more than one IP subnet over the same physical cabling.

> VLAN tagging is used on inter-switch trunks, so
> that multiple logical switches can be connected by a single physical
> circuit.  In either case, whether the switch *itself* (that is, its
> management interface) is on a particular VLAN depends on whether it
> has a level-3 address on that VLAN, at least in the normal case that
> a VLAN corresponds to an IP subnet.

The switches I have experience with-- 3com Superstack 2 and 3 family,  
and Cisco equivalents-- implement VLAN membership on a per-port basis  
via 802.1Q tagging of the packets (which can be used for hosts and not  
just switches), as well as inter-switch connectivity (what 3com calls  
VLT for Virtual LAN Trunk).

> Since we're talking here of sending IP packets, that reasoning would  
> seem to apply.  (For level 2 purposes such as spanning tree, of course  
> the switch is "on" every
> configured VLAN.)

Things aren't always as simple-- please see page 201 of the following  
document:

http://support.3com.com/infodeli/tools/switches/s_stack2/1695/ 
manual.a01/manage.pdf

"Figure 50 shows a network containing VLANs 1 and 2, and they are
connected using the 802.1Q-tagged link between Switch B and Switch C.
By default, this link has a path cost of 100 and is automatically  
blocked
because the other Switch-to-Switch connections have a path cost of 36
(18+18). This means that both VLANs are now subdivided — VLAN 1 on
Switch units A and B cannot communicate with VLAN 1 on Switch C, and
VLAN 2 on Switch units A and C cannot communicate with VLAN 2 on
Switch B.

[ ...diagram unrepresentable in ASCII... ]

To avoid any VLAN subdivision, we recommend that all connections
carrying traffic for multiple VLANs have a lower path cost than those
carrying traffic for single VLANs. You can do this in two ways:

1
 Using connections that have a higher bandwidth (which, by default,
have a lower path cost)
2
 Lowering the path cost of the connections using a Network
Management application"

>> The all-ones broadcast is supposed to go to all physically connected
>> network segments, regardless of whether a particular interface is
>> ifconfig'ured with an IP that is part of a particular layer-3 subnet.
>> You should be able to send the broadcast packet out from an interface
>> which is up but does not have an IPv4 address assigned, right?
>
> In what sense are you using "supposed" - required by some standard,
> or simply what you'd like to have happen?  If the former, please point
> to the standard.  Sending a packet out from an interface with no IP
> address assigned leads immediately to the question of what source IP
> address to use, and how a responder knows where to send its response.

They might use a source IP that starts with 169.254, per technologies  
called Rendevous at Apple, or "zero configuration" (Microsoft).

http://www.zeroconf.org has IETF drafts which discuss this, and might  
also explain my interest in bootstrapping a network without necessarily  
having host network interfaces explicitly configured with valid IPs.

> Among other things, the thing at the other end of the cable from a
> port using VLAN tagging may not approve of a frame sent
> with no tag (although cisco's can be configured with a default VLAN
> for that case).

Agreed, using a VLAN tag of 0 or 1 which represents the "default VLAN".

-- 
-Chuck