RE: [linux-tr] howto disable source route bridgeing?

[Shriman Gurung <SG@datcon.co.uk>]

Lee,

This is getting very confusing ;-))

I presume that your TR hubs have an ATM uplink to the ATM switch.  And
that the different rings connected to this switch have different Token
Ring ring numbers.  Am I OK so far?

As I understand it, if a machine M on ring A wants to talk to a machine
on N ring B, it goes like this:

1. M sends out a TEST or exchange-id (XID) frame.

This initial frame (this is layer 2) is not meant to be propagated all
over the network.  

2. M listens out for the A or C bits set in the frame, or a response.  A
is 'Address recognised', whilst C is 'frame-copied' -- indicating that
the remote machine has copied the frame into it's buffers for
processing.

3. If there has been no success, M sends another frame, but this time
includes a Routing Information Field, or RIF in the header.  The RIF
contains information on whether this frame is to be a an All Routes
Broadcast or Single-Route Broadcast as well as holding data about how
the frame gets acrodd the network (more on this later).  The choice of
ARB vs SRB lies in the network software above - netbios likes SRB, SNA
goes for ARB.

4. With ARB, each bridge that receives the frame copies it out on each
port it's connected to (save for the one it came in on, I presume).  It
adds the detail of the ring number that the frame came from and the
bridge's own number to the RIF.

5. The frame might pass through more than one bridge, each one doing the
same thing -- adding to the RIF.

6. The destination machine N receives the frame.  In fact it might
receive more than one, depending on how many different routes there are
between M and N.  In your case it sounds like there's only one route
from M to N.

7. It returns the frames to the source directly --ie not using a
broadcast, but following the 'route' specified in the RIFs of the frames
it has received. (I think it has to set another direction bit so the
frame knows it's going from N to M -- not sure though.??)

8. M receives the response(s).  It picks the route that arrives first as
the one to use for subsequent communications between M and N.  It
discards all the others.  I'm not sure this is a good idea...

4a. Single-route broadcast works slightly differently.  In SRB the
bridges 'know' the path from A to B by using the spanning-tree
algorithm.  The idea is to keep the routes short and avoid loops. (Quite
how the bridges learn the path, I've no idea)

5a. Consequently, only one copy of the frame needs to go to the
destination.  So, N gets the frame pretty much directly, courtesy of the
bridge(s),  but M, the source, still doesn't know how the frame got
there.  

6a. N sets the AllRoutes broadcast flag and sends the frame back.  The
ARB flag means that the bridge(s) will forward the frame on all
connected ports, as before.  

7a. M receives the response or responses and picks the one it gets first
as showing the route to use.


Right, so I would have thought that from reach machines on ring B from
Ring A you need to have source route bridging working, rather than
disabled.  That is, assuming the the switch makes the rings attached to
it have different numbers.  (I'm not sure about this; surely the ATM
switch doesn't know or care about SRB.)  Even if all the rings are
logically one unit, SRB shouldn't need to be disabled, since a machine
sends out a local query anyway.

Perhaps some packet traces would be useful?  The list might not be the
right place to continue this discussion (I've probably bored most
readers to sleep already, I could certainly feel myself nodding off when
I was describing SRB above :) so email me driectly if you want to
continue.

I suppose what I'd look for in a trace is what the NT machines were
doing that the linux box wasn't (or conversely).

Regards,

Shriman Gurung

System Administrator
sg n datcon n co n uk
Data Connection Ltd. -- http://www.datcon.co.uk/
--Speaking for myself not my employer--
 
PS Reengineering IBM networks by Guruge (Wiley) is fab on this stuff.
Cisco's web site is sometimes helpful too.


-----Original Message-----
From: Lee Shakespeare [mailto:lee@mbtuk.demon.co.uk]
Sent: Thursday, January 28, 1999 4:47 PM
To: linux-tr@emissary.aus-etc.com
Subject: Re: [linux-tr] howto disable source route bridgeing?


It also appears that I was blameing the wrong network component as well.
I thought that the token ring hubs where OK, but the central ATM switch
was getting confused.  On further investigation it appears that the hub
isn't forwarding the data to ATM switch.  Does this sound reasonable?
Can
someone point me to any documentation on source-routing which may help?


On Thu, 28 Jan 1999, Paul Norton wrote:

> Apples and oranges.  Token-ring source-routing is unrelated to IP
> routing.  Two different layers in the protocol stack. 
> 
> Lee Shakespeare writes:
>  > That was the way I originally thought the network would have been 
>  > configured, but alas not.  The network appears a flat class B with the
>  > hubs / switches doing the routing between each other.  
>  > 
>  > On Thu, 28 Jan 1999, Chris wrote:
>  > 
>  > > Are all the machines hanging off the rings on the same subnet?  If all
else
>  > > fails, you can set up static routes to all machines with "route"
command.
>  > > Chris
>