100 Terabit AllOptical Backbone

1
100 Terabit All-Optical Backbone

• SPAMR
• (Sam, Paul, Azita, Micah, Ramesh)
• ORS Group Project
• March 13, 2001

2
The Problem
Currently, backbone nodes of network must provide
full OEO restoration and routing In future,
magnitude of data will increase 2
Problems (A) Physical ability to route data
does not scale well (B) Cost of routing is
prohibitive
3
The Solution

• Eliminate the need for routing IF have
• Sufficient traffic between any two edge routers
  to justify NOT having to TDM/routing between them
• of edge-routers2 lt fibers wavelengths
  (available communication channels) Allows
  BUFFERLESS network by using fully connected
  mesh that is circuit switched.

4
Current Backbone
Lan/Man Edge Router Backbone Router Fiber bundle
5
Proposed Backbone
Lan/Man Edge Router Add-Drop Multiplexer
Possible switch (not router) Fiber bundle
6
Concept

• N nodes in backbone, so N2 possible connections
  (each edge-router logically connected to each of
  the others)
• Each node-node connection is on a defined
  fiber(s) at a defined wavelength(s). Propose
  100 fibers with 100 wavelengths, each channel at
  10 G bps. Total cross-section bandwidth 100 T
  bps
• At each node, the only thing that needs to be
  done is drop data incoming to that node only, add
  (on different channels) outgoing data
• Option at each node switch but not route
  channels to change fiber or wavelength --
  increase connectivity -- for provisioning and
  redundancy

7
Pros and Cons

• PRO No routing needs to be done in backbone
• Cheaper
• More scalable
• Fewer switching/routing nodes
• CON
• Potentially more complicated edge-routing
• Potentially less efficient use of network

8
Let there be light -- for a price
Use of fibers in ground is not free. Lighting
a fiber can be expensive. Potentially critical
factor in determining overall cost of
system. 50,000/EDFA 1 EDFA/30
miles 3000miles/fiber 5M 100 fibers
500M 10K fibers 50B
9
Dollar Cost of doing OEO
1 laser 10K, perhaps 1K in future 1
laser/channel 100 channels/fiber 1M/fiber
(100K/fiber in future) 100 fibers 10-100M
per node for optics alone! Expensive unless
dramatic change in technology ability to
manufacture and integrate optics and CMOS for a
few dollars instead of 100-1000? Significant
cost advantage if can avoid OEO or wavelength
switching
10
What is physically possible/SNR
QUESTION How far and through how many nodes can
a signal travel without 3R / OEO ?
Always good to improve Receivers Shot and
thermal noise, integrated detector --gt
improvement of factor of 10 --gt 10 fJ / bit
(0.1 mW at 10 GB/s) with a NOISELESS
signal Most everything else is a
tradeoff More EDFAs raise signal but add
noise Each node is loss, questions of
cascadability Longer lengths/modulator and nodes
add dispersion More wavelengths require more
EDFAs
11
Is Rapid Reprovisioning Critical? NO!?!
How much traffic between two edge-routers on
backbone? Today max is 100s of Mbps Severe
limit by routers Single wavelength channel is 10
Gbps To fill up requires 3 orders of
magnitude increase Result massive underuse of
link availability, NOT oversubscription Therefor
e Do not need connectivity (ability to handle
changing traffic flows) since most likely any two
edge-routers will not fill up a single channel
link between them, let alone several channels.
12
Which is better?
If a single channel is sufficient, is it better
for the system to have a few fibers with several
wavelengths OR a lot of fibers each with a few
wavelengths? Adding a new fiber may cost 1M to
light, But space-switching is relatively
simple and cheap (MEMS) Wavelength switching is
expensive many lasers at each switch, and
several switch nodes. Also, need more
EDFAs/fiber as wavelengths grow But
wavelength switching can potentially be
regenerating, so total EDFAs needed is
reduced.
13
More Fiber or More Lambdas?
IF we can assume that we only need a single 10
Gbit/s channel between edge-routers, does it make
more sense to have 10,000 fibers each with one or
two wavelengths or to have 100 fibers with 100
wavelengths? Recall, under this assumption there
is no switching at all in the backbone. The
number of EDFAs TOTAL is probably about the same
(a bit more for the 10000 fiber option).
Question does this mean that the overall cost
of lighting either type of network is the same?
14
More Fiber or More Lambdas(2)?
However, the major disadvantage of 100
wavelengths per fiber is that there can be much
more loss when trying to ADD/DROP a single
channel, so more sensitive components are
necessary. Assuming fiber is already laid then
the 10K fiber solution with low number of lambdas
is attractive. Assumes data to be sent
between core routers will not increase so that
more than 10Gbps will be needed in the near
future.