RE: {Possible Spam} Re: I-D ACTION:draft-andersson-mpls-g-chng-proc-00.txt

[neil.2.harrison@bt.com]

Don, you observed 07 March 2003 05:34
<snip>
> >SDH/SONET bandwidth on demand has never been deployed.  There is
> >serious question about whether this is a viable service.  The
> >technology has existed for about a decade (a little less maybe) for
> >provide ATM SVC service doing essentially the same thing.  Market
> >penetration after a decade is very close to zero for SVC.  Market
> >penetration for ATM SPVC/PVC is small and may be shrinking.  The FR
> >market seems to be shrinking as well.  FR SVC market is zero.
> >SDH/SONET does not have the bandwidth on demand capability but there
> >is ample evidence that implementing it would be a waste of time.

NH=> Mark has said effectively the same and I can confirm this.  To make
switched services work you basically need (i) very large numbers and (ii)
knowledge of the calling-rate/holding-times.  We have neither at SDH rates
or lower (ie OTN).  However, using generalised models you can show that to
had a stab at a reasonably efficient network/service-model the expected
waiting-time for a given BW 'post demand' needs to be the same order as the
holding-time of that BW.  If the expected holding-time is the order
days/weeks/months on a relatively small network population you can see that
its a nonsense to even think about going here at this point.  Note that you
can extend this arguement to the ethernet BOD case....the fundamantal
problem is the same (you can of course address niche markets in limited
locations, but not large/general-case geographical areas).

Wrt to your other observations on ATM/FR.  Yes, its been very difficult to
justify SVCs here.  This is also due to their use as product substitution
for TDM-leased-line services (to build private networks) rather than new
services.  And from what we see, the FR market is now flat but the ATM
market is still showing fairly high growth rates.
> >
> >ASON is a requirements document from ITU that assumes the SDH/SONET
> >BOND is a viable service.  This is an enormous leap of faith 
> given the
> >trends in the market.  The IETF doesn't buy into it.
> >
> This is not correct. I have had this discussion with the authors and 
> they assure me bandwidth on demand not the expected model of 
> deployment. 
> It is not precluded true but they realize the current situation.
NH=> Again I can confirm what Don is saying is correct.  We have been
consistent on this from the start.....we never saw BOD at L1 as the main
driver.  It was those driving GMPLS who were the main group arguing for BOD
at L1 to set-up 'LSPs' at will for the IP client layer.....and I can
remember arguing with them a couple of years ago on this saying 'please
don't go there'.  In fact its really hard to justify using a control-plane
at all at L1.  The benefits are minimal in our opinion (but to be fair, I
have to admit that we have some very slick NMS/OSS solutions here).
Restoration/S-PVC-like constructs are useful.  Having a control-plane
external NNI is arguably better than a NMS-plane NNI (eg less real damange
possible from any hacking).  Any control-plane solution must play
second-fiddle to the NMS.  And given operators have multiple clients, none
of which will be 'peering' with the L1/0 networks then there is no
compelling argument to select a specific control-plane solution.....the
components can therefore be chosen based on what an operator perceives to be
best-of-breed.
> 
<snip>
> >Initially IP and voice ran over TDM.
NH=> Don, they always will in core networks.  Cct-switching is a fine way to
aggregate *any* forms of large traffic volumes, irrespective of its source
nature.  This is just a fact that any operator could confirm.

regards, Neil
<snipped to end>