The MPLS-OPS Archive

[Geoff Bennett <geoff.bennett@marconi.com>]

Hi Srihari,
The simple answer is "not yet", but I need to qualify that with a bit of
background.

There is more than one optical networking architecture out there, and
currently the IP layer is simply overlayed on all of these with no
interaction between them.  So within the IP layer, we have an emerging
Class of Service (ie. aggregated behaviours, no hard guarantees)
architecture called DiffServ
(http://www.ietf.org/html.charters/diffserv-charter.html), and a QoS
architecture (per-flow or aggregated behaviour with service guarantees)
called IntServ (http://www.ietf.org/html.charters/intserv-charter.html).
Neither of these are specific to, or requiring of any underlying optical
architecture.

We can list the optical architectures of today, and the near future.

- SONET/SDH.  This is a TDM-style transmission system with hard QoS because
it does not offer any overbooking or statistical muxing.

- DWDM.  Wavelength Division Muxing also does not allow overbooking.  Any
wavelength is "service transparent", and simply gets the entire bandwidth
of the wavelength (remember, most DWDM gear doesn't even know at what
bitrate the service is operating).

- Lambda Switching (aka GMPLS).  The current proposals essentially describe
the way that an MPLS-like control plane can be used to set up
service-transparent DWDM channel trails.  So there is no difference in QoS
terms between DWDM and Lambda Switching.  HOWEVER, because GMPLS requires
that MPLS LSPs are aggregated into an optical channel trail, there does
need to be a crude interchange of QoS information between the optical layer
and the packet layer.  But all the proposals for this function involve
electronic grooming devices at the edge of the network.  The core of the
network is still "service transparent".  Bernstein and Sharma give a brief
overview of the parameter interchange and implications for routing
(http://search.ietf.org/internet-drafts/draft-bernstein-gmpls-optical-00.txt
), but this stuff is at a very early stage right now.

So the current, and nearly-deployed optical architectures are "mostly
decoupled" from IP-style service quality.  Let's look further in the future...

- The IEEE is developing 10G Ethernet as a ring-based metro network
technology.  There is a study group in IEEE that is developing a spatial
reuse protocol for ring architectures to allow them to break away from the
provisioning limitations of TDM
(http://grouper.ieee.org/groups/802/rprsg/public/presentations/may2000/).
But these efforts will probably result in hybrid opto/electronic devices,
where the muxing is done in the electronic part of the box, and this will
limit transmission speeds to the 10G range in the near term.


For high speed optical operation we need to have true optical switching,
and there are at least two broad thrusts under way right now; optical
packet switching and optical burst switching.  Both of these techniques are
focussed on solving two really hard problems; first how do you do
statistical muxing in all-optical components; second, if we are
transmitting packets at 40Gbps or faster, how on Earth do we build
electronics that can interpret packet headers to read addresses and labels?


Buffering: to do statistical muxing at the packet level in all-optical
devices is quite hard.  This is especially true if you allow variable
length packets.  But the biggest problem is that it is currently very, VERY
difficult to built optical memory for buffering in a switch.  In the past,
all packet switches have needed buffering; on the input side you have a
tiny little buffer so you can look at the header of the packet to interpret
addresses or labels.  On the output side you need big buffers to deal with
contention.  You never build an INPUT-buffered switch (neither electronic
nor optical) because you'll end up with Head-of-Line blocking, which is a
bad thing if you're intending to implement a QoS architecture.  Currently
the only practical optical buffers are lengths of fibre.  In other words,
if you have contention on the output port of an optical switch, throw one
of the packets into a few kilometres of fibre to let the queue clear.
Imagine a switch with 1,000 ports with several kilometers of fibre on each
one!!!

Optical Packet Switches have to have buffers.  This is a big disadvantage.
There is a way to build Optical Burst Switches without buffers, and still
get statistical muxing.

High Speed Bit Reading:  Optical Packet Switches are designed to transmit
packet headers at a slower speed than the packet payload.  Strange but
true.  Optical Burst Switches use an out of band signalling technique to
establish channel trails, and this out of band channel can run at a lower
bit rate.

Both OPS and OBS are in the reasearch phase.  Checkout
(http://www.cse.buffalo.edu/~qiao/) for OBS.  Professor Ivan Andonovic at
Strathclyde University (http://www.strath.ac.uk) is doing work on OPS.

Cheers,
Geoff


At 19:02 25/11/00 -0800, sraghava wrote:
>Hi all,
>Are there any QoS-architectures for optical networks (say, IP over DWDM)?
>What are the Qos-parameters for optical network? Is there any MPLS based QoS
>architecture for optical networks?
>Any help is greatly appreciated.
>
>Thank you,
>
>Srihari Raghavan
>CS Graduate Student
>Virginia Tech
>home: 540 552 6397
>===================
>
>-------
>The MPLS-OPS Mailing List
>Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
>Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
> 
================================================================
Geoff Bennett                             Tel: (33) 497 21 43 62
Director, Office of the CTO               Fax: (33) 497 21 43 50
Marconi
Gaia - Bat. E                           email: geoff.bennett@marconi.com
BP 123
06903 SOPHIA ANTIPOLIS
FRANCE
================================================================

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml