Original Question:
>I've just started a new job and the routing in my lab seems a little
>different than what I'm used to:
>
> 1) Machines don't set a default route to a gateway.
> 2) They don't run routed
> 3) They DO set a default route through themselves with a hop
> count of 0. eg:
>
> route add default `uname -n` 0
>
>Most machines are running Solaris 2.[45]. The rest of networking
>config is as follows:
>
> class B network with subnetting (255.255.255.0)
> Using a "0" broadcast address.
>
>I've never seen anything like this before. My gut reaction is that
>this is "broken" but I can be taught :^) A read of the route man page
>indicates that this setting causes the client station to use ARP to
>resolve the addresses. The gateway would then respond (provided its
>using RIP and not OSPF).
>
>Can someone please explain to me why this works? Thanks - I'll
>summarize.
>
Answer:
The short answer is "Yes". A hop count of zero causes the client to
ARP for the destination address. A gateway running proxy arp should
send the packet to its intended destination.
What seems to have happened during the gateway conversion is that
proxy arp was turned off, breaking the implemented routing scheme.
We've since implemented the "standard" routing technique of setting a
default router and things are fine.
Special thanks to the following:
Luca Pizzinato (pizzinato@eumetsat.de)
Mike Salehi (mrs@cadem.mc.xerox.com)
Matthew Stier (mstier@hotmail.com)
Andrew Mellanby (mel@maths.abdn.ac.uk)
Dave Haut (dhaut@level1.com)
More complete answers follow:
>> From: Luca Pizzinato (pizzinato@eumetsat.de)
Hi.
Yes, it is a valid network configuration. I configured a
department with HP clients just like that.
You are right, when a client has to send a packet to another
client, first it reads the ARP table. If the destination is not
there, it issues an ARP request, otherwise it's not necessary
to do it. An ARP entry is automatically deleted only after
20 minutes (default, if the kernel of your clients have not
been modified), your clients don't issue an ARP request for
each IP packet to be delivered.
If you say that the broadcast has been set as 0, I think that
all clients are on the same LAN (no routers inside), so with
such broadcast you are sure that ALL clients will receive
the Ethernet frame, and only the destination client will reply.
In my case, I have all subnet connected to a router running
ProxyARP, so each client sends the ARP request out to his
local network and the router answer, playing the role of the
destination client. The sending client thinks that the router is
the destination and sends the packet to it, leaving to the
router the problem of delivering it.
In other words, if your subnets are connected to routers,
then you can set the broadcast as xxx.xxx.255.255, relying
to routers the delivery.
If there are still points to be cleared or discussed, let's keep
in contact.
Have a nice day
Luca Pizzinato
pizzinato@eumetsat.de
Tel. +49.6151.807323
Fax +49.6151.807304
>> From: Mike Salehi (mrs@cadem.mc.xerox.com)
John,
I was miffed by it, and I never found any documentaion of why it works.
But I think its a BSD convention when you route like that you send packets
on the
ether and its the routers responsibility to listen to packets and route
them properly.
The RIP packets will go the same way and the router will pick it up and
broadcast it
to other routers. That's how I theorized and if you find a definitive
answer please let me
know too.
>> From: Matthew Stier (mstier@hotmail.com)
I've had network admins tell me that this will work.
I've always gone with /etc/defaultrouter (if there was only one way off the
network) or rdisc (router discovery) or rip dependent upon the routers.
-- Matthew Stier mstier@hotmail.com--------------------------------------------------------- Get Your *Web-Based* Free Email at http://www.hotmail.com ---------------------------------------------------------
>> From: Andrew Mellanby (mel@maths.abdn.ac.uk) At 11:35 15/04/97 -0400, you wrote: >Greetings, > >I've just started a new job and the routing in my lab seems a little >different than what I'm used to: > > 1) Machines don't set a default route to a gateway.
That's OK if you don't want them to talk to other networks
> 2) They don't run routed
Only routers (ie. machines with more than one interface) need to run routed - even then, static routes can usually do the job.
> 3) They DO set a default route through themselves with a hop > count of 0. eg: > > route add default `uname -n` 0
I don't see the point.
>Most machines are running Solaris 2.[45]. The rest of networking >config is as follows: > > class B network with subnetting (255.255.255.0) > Using a "0" broadcast address.
x.x.x.0 is the old way of doing it, should be x.x.x.255 (maybe they read an old book)
>I've never seen anything like this before. My gut reaction is that >this is "broken"
If it works it isn't broken. I would certainly try to sort it out though.
---- Andrew Mellanby Computing Officer, Dept. Math.Sciences, University of Aberdeen http://www.maths.abdn.ac.uk/~mel (All views expressed are my own, NOT those of the University)
>> From: Dave Haut (dhaut@level1.com) > > Greetings, > > I've just started a new job and the routing in my lab seems a little > different than what I'm used to: > > 1) Machines don't set a default route to a gateway. > 2) They don't run routed > 3) They DO set a default route through themselves with a hop > count of 0. eg: > > route add default `uname -n` 0
This is totally valid. What this does is force the machine to send ANY packet, regardless of it's ip address through it's own interface without sending it to a router first. This is why you don't need routed or a default route setup for it ...
Hope This Helps ...
--------------------------------- _ /| Dave Haut \,o.O' Supervisor, Sys Admin Group =(___)= Level One Communications U dhaut@level1.com (916) 854-2838 ---------------------------------
-- John M. Vogtle Phone: (716) 477-3045 Senior UNIX Administrator US Mail: Eastman Kodak Company jmvogtle@kodak.com 6/69/RL MS: 01925 One planet is all you get. Rochester, New York 14650-1925
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:55 CDT