Cell Relay Archive

[manfredi@arl.bna.boeing.com]

In article <33D5FE90.23D1@tcom.epfl.ch>,
  Jean-Philippe Martin-Flatin <jpmf@tcom.epfl.ch> wrote:
>
> Hi,
>
> We are currently integrating connection admission control (CAC) modules
> with a pseudo expert system which periodically (period T) computes the
> optimal routing tables for a whole ATM network, based on statistics
> retrieved from the CAC on each node of the network. In this network, all
> VC and VP connections are VBR. The bandwidth of each VP is variable (UNI
> 4.0), and optimized periodically by the pseudo expert system. What we
> model here is a service provider trying to reduce its bill to the
> network operators.
>
> If we assume that all VC connections in our network are short-lived,
> then the problem is simple: as soon as the routing tables are updated,
> connections are created using the new routes, and soon afterwards all
> connections created prior to the routing table update are released.

I would have worded that differently. I would have said that new VCs set
up after the routing tables are updated would use the new routes. But VCs
which existed before the routing table update would _not_ be torn down
automatically. At least, that's not the normal way of doing business. (I
suppose one could implement a special-purpose ATM network that way, to
somehow force VCs to be torn down and re-established whenever a routing
table change occurs.)

> Therefore, in the expert system computing the optimal routing table, we
> can safely assume that we start from scratch every T, and neglect
> existing connections.

In a model, that might be close enough, assuming short-lived VCs.

> If we assume that some VC connections in our network are long-lived, but
> the aggregated bandwidth taken up by all these connections is very small
> compared to the VP bandwidth, we can just ignore them in our expert
> system. But if this aggregated bandwidth is relatively large, we can no
> longer neglect them: this makes our routing optimization algorithm
> substantially more complex. We are precisely interested in the latter
> case.
>
> We have 2 possible scenarios: either we consider that all existing VC
> connections are static (cells go across the same ATM switches throughout
> the life of the connection), or we consider that they can be rerouted on
> the fly, transparently, according to the new routing tables. The
> question is: does it make sense to consider the rerouting of these
> existing long-lived, bandwidth demanding VC connections ? Is it
> something made possible at all by ATM ?

Re-routing of calls, if you use Q.2931, would only happen if the VC is
first torn down, then re-established. This is what makes circuit-switched
and packet- switched systems different (ATM being circuit-switched).

> If so, is it something done in
> practice, or at least prototyped, or is it a purely academic case ? I'm
> not sure what spec I should read to get the answer to this. I checked
> UNI 4.0 and the ATM FAQ, but didn't find anything relevant. I know
> little about SS7, is it there ?

>From what you say, I think you should read the PNNI document:
af-pnni-0055.000, available at the atmforum FTP site.

Bert
manfredi@arl.bna.boeing.com

-------------------==== Posted via Deja News ====-----------------------
      http://www.dejanews.com/     Search, Read, Post to Usenet