RE: draft-fedyk-gmpls-ethernet-pbt-00.txt

[Dimitri.Papadimitriou@alcatel.be]

hi don

when you write

> BTW centralized path control may need a PCE system not necessarily a 
management system. 

Path Control and Path Computation - are two different functions - 

i know that the acronym PCx could be misleading but hope not until that 
point

thanks;
- dimitri.





"Don Fedyk" <dwfedyk@nortel.com>
Sent by: owner-ccamp@ops.ietf.org
30/06/2006 21:44
 
        To:     "Lucy Yong" <lucyyong@huawei.com>, "Igor Bryskin" 
<ibryskin@movaz.com>, <ccamp@ops.ietf.org>, <gels@rtg.ietf.org>
        cc: 
        Subject:        RE: draft-fedyk-gmpls-ethernet-pbt-00.txt


Hi Lucy 

<snip>
> > 
> > Which model do you pursue for PBT, why?
> Either model is appropriate. It comes down to a Network 
> providers requirements to manage the network and the 
> resources. For many reasons initial PBT services are static 
> from a source/destination view point.
> In some cases we will be starting from a centralized 
> management system.
> However, the aspects of GMPLS signaling and control allow a 
> range of static to dynamic paths even though the endpoints 
> are static. The real issues related to dynamic control are 
> much like L1VPNs and some of Igor's comments.  If the service 
> has an identifiable UNI a more dynamic model may be developed 
> but right now we specifying the provider operations. 
> 
> [Lucy] On demand request from customer space could be 
> implemented in two ways also, embedded network intelligence 
> or management plane driven.
> Somehow, I think this is the architecture design issue not 
> strictly related to on demand request from customer. 
> Basically, how we want to build an end to end forwarding 
> path, rely on network intelligence or network management system?

Let me paraphrase your comment. Does the network have the intelligence
for path creation or does the intelligent lie in the management system? 
I assume by network intelligence you meant he typical TE database? 
Well, this is just my opinion, we know when path holding times are long
and network fill is a concern that centralized management systems can
add extra policy type operations that we do not have in today's typical
TE network database. 
We also know that when path turnaround times are short or there is ample
bandwidth that any reasonable path will do network intelligence of today
TE database is quite adequate. This leads me to believe that initial
path requests will have  more centralized path control but path
maintenance operations are logically handled be the network (eg
rerouting, reversion etc). BTW centralized path control  may need a PCE
system not necessarily a management system. 

Of course the other interpretation of your question is do we augment the
network with more intelligence? In some ways we proposed to do this for
optical systems but uptake on augmenting the TE database has been slow
going. 

I don't think that PBT is special in this regard I think that generally
these questions are applicable to any of the GMPLS systems. 

Regards,
Don 
> 
> Regards,
> Don 
> 
> > 
> > Best Regards,
> > Lucy Yong
> > 
> 
> 
> 
> 
> 
>