Title: Optical Switching Networks
1Optical Switching Networks
- Presentation by
- Joaquin Carbonara
2References
- Work by
- Ngo,Qiao,Pan, Anand, Yang
- Chu/Liu/Zhang
- Pippinger/Feldman/Friedman
- Winkler/Haxell/Rasala/Wilfong
3Introduction
4About Optical Networks
- Wavelength-routed all-optical WDM networks are
considered to be candidates for the next
generation wide-area Backbone networks
Chlamtac,Ganz and Karmi, 1992 and Mukherjee,
2000 - Wavelength Routed Network wavelength routers
connected by fiber links (each being able to
support wavelength channels by supporting WDM) - WXC can be uni/multicast. OXC can be used between
processors in a parallel or distributed system.
5About Optical networks
- In a dynamic wavelength-routed WDM network,
limitations of the network may result in some
light-paths requests not being satisfied. - Goal design all-optical networks that minimizes
blocking.
6About Optical Networks
- Wavelength Continuity Constraint (which makes
Optical nets different than circuit-switched
telephone nets) Thus two light paths that share
a common fiber link should not be assigned the
same wavelength. - Solution Wavelength converters.
7About Optical Networks
- Switching speed is the bottleneck at the core of
the optical network infrastructure Singhal and
Jain, 2002 - Goal design cost-effective WXC that are fast and
easily scalable.
8Design analysis
- RNB (rearrangeable non-blocking) a set of
requests submitted at once can be satisfied by
the network. - SNB (Strictly non-blocking) a new request can be
satisfied without changing current request paths. - WSNB (Wide sense non-blocking) a new request can
be satisfied using an (on-line) algorithm. - SNB --gt WSNO --gt RNB
9Design Analysis
- Cost of components is important.
- Number of different components
- (de)multiplexors (MUX/DEMUX)
- Wavelenght converters (full-FWC or limited-LWC)
- Semiconductor Optical Amplifiers (SOA)
- Optical add-drop multiplexors (OADM)
- Arrayed Waveguide Grating Routers (AWGR)
10Design Analysis
- Theoretical results help understand and design
networks - Complexity is important (as a function of size)
- Size number of edges in graph theoretical
representation - Depth number of edges in longest path of graph
theoretical representation.
11Design tools
- Mathematical modeling
- Graph Theory Theory of Discrete
Mathematics/Combinatorics Functions
(Real/Integer valued, one or more variables)
Linear/Multilinear Algebra. - In mathematics you don't understand things. You
just get used to them. - von Neumann, Johann (1903 -
1957) - Mathematicians are a species of Frenchmen if you
say something to them they translate it into
their own language and presto! it is something
entirely different. - Goethe (German writer),
Maxims and Reflexions, (1829)
12Design tools
- Advantages of mathematical modeling
- Many tools available since Mathematics is an old
and well established discipline - True statements are backed by proofs (100
guaranteed--if used properly). - Math language is practically universal. This
guarantees a larger audience . - Math organizes knowledge extremely well.
13Design tools
- Disadvantages of Mathematical modeling
- It is hard to fit reality into a nice Theory
- Theory requires organized abstract thinking--not
a very popular activity
14Design Tools
- Other tools include simulation and analysis (I
will not talk about these tools).
15Optical Network Design
- Definitions, Examples and Theoretical Results
16Heterogeneous WDM Cross-Connect
17Components Wavelength Converters
- Wavelength converters take as input wavelengths
coming on different fibers and can be programmed
to modify the wavelength and output modified
wavelength. - To reduce cost, researchers have
- Used Limited Range Wavelength converters (LWC)
instead of Full Range Wavelength converters (FWC) - Share wavelength converters among fiber links.
- Notation LWC(A,B) takes inputs from set A and
produces outputs from set B.
18ComponentsAWGR
- Arrayed Waveguide Grating Routers
- Passive devices reroute channels inside fibers
- Easily available and inexpensive
- Take m inputs and have m outputs fibers
- Process wavelengths 0 to m-1
- Wavelength i at input fiber j gets routed to the
same wavelength at output fiber (i-j)mod m.
19Request Model(understanding Nets blocking
properties)
- Model 1 -- (?, F, F?) Requests are of the form
(?i, Fj, F?j ? ) where ?i is a wavelength, Fj is
an input fiber and F?j ? is an output fiber.
Requests requires only an given output fiber, but
do not specify the output wavelength. - Model 2 -- (?, F, ??, F?) More restrictive than
Model 1 since output wavelength is also
requested. - Note If N satisfies M2 then it satisfies M1
20WXC-RNB construction for M1(Ngo/Pan/Qiao infocom
04)
- Components Let f2, b3, n4. Then it has
- f demultx, fbn LWC(Bi,n), fb n ? n-AWG,
- fbn LWC(n,bc,b(fc)), n multx,
- nb ? nb-AWG, and f multx.
21WXC-RNB-1 means ...
- RNB means that any set S of valid requests will
not be blocked in the network N. While in transit
inside the network, the Wavelength Continuity
Constrain must be satisfied. - Valid request means
- no two requests will ask for the same input
wavelength and fiber. - the number of requests asking for the same output
fiber cannot exceed the fiber capacity.
22WXC-RNB-1 and GT
- Konigs 1916 Theorem Let G(U,VE) be a bipartite
graph. Then the maximum (vertex) degree equals
the chromatic index. - Chromatic index minimum number of colors needed
to edge color G so that adjacent edges use
different colors.
23About Konigs Theorem
24Back to WXC-RNB-1...
- Represent the network as a bipartite graph
G(U,VE) for the sole purpose of determining a
non-blocking route for each request - The set U corresponds to the set of input bands
(there are fb of them) - The set V corresponds to the set of output fibers
(there are f of them)
25Graph of WXC-RNB-1
- Represent the network as a bipartite graph
G(U,VE) - Request (?p, Fq, Fj) ? edge (ui,vj)
- where i qb ?p/n?
- By a simple variation of Konigs theorem, the
graph G is colorable with n x b colors (label
each color with a tuple (c,d)), 1c n and 1d
b, in such a way that edges sharing a vertex in
U have different first color component.
26Routing in WXC-RNB-1
- The basic idea is this
- 1. request (?p, Fq, Fj) ? edge (ui,vj) ?
color (c,d) - 2. Then Route ?p so that it ends up in the cth
output line of its stage-1 AWGR. - 3. Working from the other end, we want the
request to end in Fj. There are b fibers demuxing
to it. We can see that if the stage-2 LWC routes
the wavelength to its dth line of its demuxer,
the desired output is obtained. -
27Routing in WXC-RNB-1
- The basic idea is this (cont.)
- 4. The properties of the coloring inherited from
Konigs theorem guaranteed non-blockiness. -
28Example (Ngo/Pan/Qiao) WXC-RNB in Model-2
29Other interesting results related to non-blocking
networks
- Strictly non-blocking networks are highly
desirable. It is difficult to build such networks
that are cost efficient. - An interesting result (Ngo)
- WXC-SNB-1 if and only if WXC-SNB-2
30Haxel/Rasala/Wilfong/Winklers work on WDM
Cross-connects
31On the news...
32Optical Network Complexity
- Graph Theoretical representations, Bounds,
minimizing the number of components. Examples and
theoretical results.
33Complexity Minimizing the Number of LWC
- Results related to using the least possible
number of LWC on a uni/multicast network - Define LWC(d) when LWC can convert ?i to ?j iff
i-jd. - Consider Homogenous Model-2 of requests with w
wavelengths and f fibers (HM2(w, f)). - Want to study statistic m1(w,f,d) least number
of LWC(d) needed if HM2(w, f) is SNB.
34Complexity of WDM networks(unicast) m1(w,f,d)
even w (Ngo/Pan/Yang)
35Complexity of WDM networks (unicast) m1(w,f,d)
odd w (Ngo/Pan/Yang)
36Complexity Size and Depth using GT
representation
- (Ngo) Using the DAG model (Directed Acyclic
Graphs) we can establish a formal definition of
size and depth of a network. - Size number of edges in the graph
- Depth number of edges in the longest path.
37ComplexityUsing Graph/Theoretical Representation
- (Ngo) Graph Theoretical representation.
- a) Fiber-channels get replaced by vertices
- b) Edges capacity
38ComplexityUsing Graph/Theoretical
RepresentationExample
- Size of the network is number of edges.
- Depth is longest path.
- It uses 2 2x2 AWG, 4 FWC 2 multiplexors and 2
demultiplexors - DAGDirected Acyclic Graph
39Graph/Theoretical Representation(Winkler/Haxell/R
asala/Wilfong)Dynamic bipartite graphs
40ComplexityUsing DAG GT RepresentationRigorous
Setting Model-2
- DAG model networks as follows
- (n1,n2)-network is a DAG N(V,EA,B)
- Vvertices, Eedges, Ainputs, Boutputs,
- n1 A, n2B.
- We can now define request, request frame, route,
RNB/SNB/WSNB network. - Key idea requests path must be disjoint to be
(simultaneously) realizable.
41ComplexityUsing Graph/Theoretical
RepresentationRigorous Setting Model-1
- DAG model networks as follows
- w,f-network is a DAG N(V,EA,B)
- Vvertices, Eedges, Ainputs, Boutputs
- ABwf and BB1 B2 ... Bf
- We can now define request, request frame, route,
and RNB/SNB/WSNB w,f-network.
42Complexity DAG size
- Let an n-network be a Homogeneous Network with n
inputs and outputs. If the output is further
divided into f bands of size w (needed for M-2)
we call it a w,f-network. - The smallest number of edges (size) for it to be
SNB, RNB, and WSNB is sc2(w,f), rc2(w,f) and
wc2(w,f) (Model 2), or sc1(n), rc1(n) and wc1(n)
(Model 2). - rc1(n)wc1(n) sc1(n),
43ComplexityResults from DAG model
- M1 is less restrictive than M2 since M2 requests
specify an output wavelength. The following
result shows that in the SNB case there is no
difference in cost between models
44ComplexityRNB w,f-networks
- The size function has known estimates in this
case
45Complexity
- Advantages of having bounds
- Number of edges can be related to network cost
- Theoretical results are often the only way to
gain experience with abstract systems. Examples
may be too poor or difficult to concoct.
46Complexity
- Other results include different ways of create
atomic networks, and operations to create
larger networks from smalles - Left and right union
- The ??-product
47Future Work
- Expansion of current models using different
models with the goal of eliminating blockiness
while reducing cost. - Search for better bounds on the current
statistics. - Search for new meaningful statistics (is size and
depth the only ones that matter?) on GT
representations.