On 19 Sep 2003 17:28:23 -0500
Bret Hughes <[EMAIL PROTECTED]> wrote:
> I have zero experience with wireless so excuse my boneheaded
> question(s). I think this is going to make me take an orthogonal view
> from how I normally picture a network. Perhaps it is my needing to
> read
> up on routing daemons that would fill in my blank spots.
Hey Bret,
I have only slightly more experience with wireless networks and because
of the complexity was only trying to nudge the OP in the right direction
rather than provide a solution.
The solution i did hint at wasn't meant specifically for wireless, same
thing can be setup on a local lan. For instance in a classroom
situation each node is configured with multiple logical networks on the
same physical lan. Even though all the packets are flying across the
same physical media, only the proper network "sees" each packet because
of the logical segmentation. This was the genesis of the notion that i
described.
AFAIK bridging isn't an option on wireless, at least not for this kind
of relay network, if it were then it would be possible to treat all the
nodes as a single segment with no need for layer 3 routing. But then
again the subject line did suggest a routing question not a wireless
question ;o)
>
> My usual thinking ( forget wifi for a minute)
>
> base server --------node1---------node2 ... node12-------remote
> server
>
> in a wired scenario each node would have 2 nics and each segment
> between the boxes would be separate networks with exactly 2 hosts.
> 4 ipaddresses would be required to correctly subnet each network.
>
> Now, Sean said make each node a network and since there is only one
> interface that can talk to multiple networks, this is where I get
> bumfuzzled since as I said, I usually thing of the link between
> machines as the network.
>From each nodes perspective it is on a wireless lan consisting of just
its immediate neighbors. The problem becomes how to route packets
between these smaller overlapping physical networks. Ideally a
solution would route packets as far as possible in each hop and be
robust enough to adapt if a node goes down.
example:
All of the nodes are on the same wireless subnet. However each node is
also configured internally with it's own unique network. For example
node4 also responds to IP's 10.10.4.1 and is the only node on the
10.10.4.0/24 network. (I suspect this is the point that wasn't clear
in my original post)
wireless lan: 192.168.0.[1-5] one for each node
unique network IP: 10.10.[1-5].1 one for each node
base1-------+---------+
node2-------+----------+
node3-------+---------+
node4------+
remote5
(diagram sux unless viewed with a fixed width font)
base1 routing table:
dest gateway
10.10.2.1 192.168.0.2
10.10.3.1 192.168.0.3
10.10.4.1 192.168.0.3
10.10.5.1 192.168.0.3
node3 routing table:
dest gateway
10.10.1.1 192.168.0.1
10.10.2.1 192.168.0.2
10.10.4.1 192.168.0.4
10.10.5.1 192.168.0.5
When base1 wants to send a packet to node2 it _could_ use 192.168.0.2
directly because that node is in wireless range. But what if it wants
to reach remote5? 192.168.0.5 is NOT in wireless range.
So instead of sending to 192.168.0.5 we send to IP address 10.10.5.1.
At this point the routing tables take over and forwards the packet to
node3 (192.168.0.3). Node3 has a direct route entry so forwards the
packet again to 192.168.0.5 (from here 192.168.0.5 is in range!) When
the packet arrives at 192.168.0.5 the node recognizes that requests for
10.10.5.1 are for itself and processes it locally.
So the answer is to _always_ use the "fake" network/node number rather
than the real wireless lan IP, routing will take care of the rest.
While you could do this all with static routes it's not as flexible.
So the only remaining setup is to use a routing daemon to configure the
routing tables so that they are adaptable to outages and the like. On
a good wireless day with little interference perhaps the routing tables
will show that node4 can be reached directly from base1 without a need
to route through node3. Other days packets destined for node4 from
base1 will have to route through both node2 and node3.
>
> I think of this as logical links (assume that each machine can see the
> next two)
>
> /---------------- ---\
> base server --------node1---------node2 ... node12-------remote
> server
> \------------------------------/
>
> Since the network interface on any given node can see multiple
> networks is this tantamount to running multiple subnets on the same
> wire? Is there the concept of a default gateway in this scenario? I
> can't see how since in all neworking I have done the next hop or
> default gateway to another network is on the same subnet as the
> host.
>
No need for a default gateway, although it doesn't hurt to have one.
The default gateway is only used when a packet to be routed doesn't
match the destination of an explicit route.
> Does this make sense? what am I missing?
>
Only a bunch of important parts that i left out. ;o)
Cheers,
Sean
--
redhat-list mailing list
unsubscribe mailto:[EMAIL PROTECTED]
https://www.redhat.com/mailman/listinfo/redhat-list
