Cell Relay Archive

[andy@legion.demon.co.uk (Andy Leigh)]

Following on from Fahad's post regarding how VPI/VCI pairs are
selected on a per-hop basis, I got the feeling that the original
poster wanted to know how to build a big WAN.

Each ATM device has a unique (due to a combination of 6-byte
manufacturer's sig + organisational additions) ATM address. The ATM
address has sections that can be used by end-companies to register a
range of addresses uniquely, and then sub-net internally (just like
IP, only much, much bigger).

Therefore, any end station (software) needs only to know what the end
ATM address that it wants to get to. This is the same as IP (e.g. type
"telnet 192.158.30.72") or like a phone number. How can you know this
address? You need a directory, like a phone book or DNS...

The end station requests of the switch it's connected to, "please
connect me to ATM address: X". The switch has to know how to get to
this address. How is that done? Either via static routing (similar to
setting up default routes in IP) or via dynamic routing protocols. In
most cases this latter is P-NNI (or B-ISUP). What's P-NNI? Well it's
OSPF on steriods, capable of > 100 levels of information hiding
(compare this with 2 for OSPF!), as well as aggregating QOS
availability across the network (I wait to see how useful this will be
in practice). Thus the switch knows the best path to forward the
request to. It opens VCI/VPI pairs (as Farad explained), until you get
to the end destination. Piece of cake!

Andy

fahad@cs.pitt.edu (Fahad A Hoymany) wrote:

> JoongSub Lee (kornet) (audience@soback.kornet.nm.kr) wrote:
> : Hi,
> 
> :  I know this qestion is so silly but I can't proceed to learn
> : ATM because of this question.
> :  
> : With my short knowledge it is impossible to cover WAN network
> : with VPI/VCI. If A want to communicate with B, how routing
> : is made? In TCP/IP the destionation address never change so they
> : can match it eventually but ATM change when transit switches.
> : I understand communication is set-up after routing decison making
> : via VPI/VCI. But I don't know how routing decision is made when
> : virtual circuit is set up.
> : Thanks in advance.
> 
>  Lee,
> 
>     The following is an example I made some time ago... it should answer
> your question or part of it.
> 
>   Lets say we have user A wishing to connect to user B across an ATM
> network:
> 
>                   Switch 1           Switch 2
>                 +--------+         +--------+
>                 |        |         |        |
>    A ___________|        |_________|        |_____________ B
>                 |        |         |        |
>                 +--------+         +--------+
> 
> 
>   'A' sends a setup message without any VPI/VCI values as we all agreed on,
> whether ATM Forum or ITU. Switch 1 selects a VPI = 45, VCI = 27 (as an
> example) and sends it back to A in its CALL-PROCEEDING message.  So, we have
> this picture:
> 
>                   Switch 1               Switch 2
>                 +--------------+         +--------+
>                 |  in     out  |         |        |
>    A ___________|45 27 a       |_________|        |_____________ B
>                a|              |b       a|        |b
>                 +--------------+         +--------+
> 
>   Note that switch 1 does not have a value for the out direction yet. 'a'
> and 'b' are the port numbers. Switch 2 then selects a value (37,95) and lets
> assume that the NNI protocol requires switch 2 to send a CALL-PROCEEDING
> message back to A. Then A updates its table as follows:
> 
>                   Switch 1               Switch 2
>                 +----------------+         +--------------+
>                 |  in     out    |         |  in      out |
>    A ___________|45 27 a 37 95 b |_________|37 95 a       |_____________ B
>                a|                |b       a|              |b
>                 +----------------+         +--------------+
> 
> 
>   What this so far says is that any cells coming from A marked 45,27 should
> go out on port b as 37,95. The trouble arose on whether switch 2 selects
> the remaining entry or user B selects it. The UNI 3.1's method as I learned
> today selects some values and sends them to B.  If B cannot accept them the
> connection fails (see section 5.5.2.3). The method I suppose suggested by
> ITU lets B select its own numbers and return them in its CALL-PROCEEDING
> or CONNECT message.  In this case, we would have something like:
> 
>                   Switch 1               Switch 2
>                 +----------------+         +----------------+
>                 |  in     out    |         |  in      out   |
>    A ___________|45 27 a 37 95 b |_________|37 95 a 87 57 b |_____________ B
>                a|                |b       a|                |b       87,57
>                 +----------------+         +----------------+
> 
> 
>   So, now any cells out of 'A' marked 45,27 go out of switch 1 as 37,95 on
> port 'b'. Cells coming to switch 2 with 37,95 go out as 87,57 on port 'b'.
> (Of course I am simplifying the discussion here by showing only one way
> of the communication and I am not worried about the exact nature of the
> call establishment protocol)
> 
> Fahad