rev pa1 tcom rev pa1 1/37 tcom optical networking: principles, hot topics and future perspectives...
TRANSCRIPT
Rev PA1
TCOM
Rev PA1 1/37
TCOM
Optical Networking: Optical Networking: Principles, Hot Topics Principles, Hot Topics
and Future Perspectivesand Future Perspectives
November 24November 24
Sébastien RumleySébastien Rumley
EPFL – EPFL – Laboratoire de Laboratoire de Telecommunication Telecommunication (TCOM)(TCOM)
[email protected]@epfl.ch
TCOM
Rev PA1
TCOM
Rev PA1 2/37
TCOM
Outline• 1. Optical networking : principles
– Optical transmission (history, rules of thumbs, etc.)
– Optical switching
– Context and challenges
• 2. Research fields– Components
– Network Management schemes
– Optical networks design and planning• Routing and Wavelength Assignment (RWA)• Routing and Regenerator Placement (RRP)• Optical Burst Switching
• 3. Alternative ideas– All Optical Network Coding
– OBS multicast
– Ad-hoc wireless optical networks
Technological research
Academic research
Rev PA1
TCOM
Rev PA1 3/37
TCOM
Principles : Optical transmission
• An optical fibre is a wave guide.
• The light emitted by a laser diode is modulated and injected in this guide
• The modulated signal is analysed with a photodiode at the fibre end
Fibre (wave guide)
Laser + modulation
Input data
Photodiode
Output data
• The transmitted signal is subject to
– Attenuation (bending, absorption, scattering)
– Impairments (chromatic dispersion, polarisation, non-linear effects…)
Impairments Noise
Attenuation + noise Signal to Noise Ratio (SNR) !
Absorption/ScatteringBending losses
Rev PA1
TCOM
Rev PA1 4/37
TCOM
Raw transmission limits
1. The attenuation is frequency dependent– A fibre is a very good wave guide for given frequencies only
– Typically in three “windows” located around 1.4 μm– Outside, light is attenuated by various effects
2. Modulation speeds– In theory, bandwidth of ~2x20 Thz
~80 TBaud/s (Nyquist)– Not realisable in practice
3. Impairments:– Non linear effects– Chromatic dispersion
Image source : http://www.fiberoptics4sale.com/wordpress/optical-fiber-attenuation/
This profile is different for each fibre
Rev PA1
TCOM
Rev PA1 5/37
TCOM
Transmission enhancement : multiplexing
• Multiplexing of independent channels (λ), modulated at lower rates WDM (Wavelength Division Multiplexing)
Good news :• Capacity multiplied
Bad news :• Crosstalk
Noise
• Demultiplexing losses Attenuation
Image sources : http://www.imec.be/ScientificReport/SR2008/HTML/1225202.htmlhttp://www.harzoptics.de/pof-demultiplexer.html
Rev PA1
TCOM
Rev PA1 6/37
TCOM
Trans. enhancement : optical amplification
• How to mitigate the attenuation?• How to carry signals over longer distances?
By reamplifying them !
• Electronic amplificationDemodulate the light signal with a photodiode, remodulate a new signal
+ Not only amplification (Resizing), but also Retiming and Reshaping (3R)- Resource and energy consuming- With WDM, demultplexing and remultiplexing is required- One regeneration per signal
• Optical amplificationSimilar principle as the laser
+ All channels amplified simultaneously- Amplification only- Works only for given frequencies
Rev PA1
TCOM
Rev PA1 7/37
TCOM
Historical landmarks for transmission
• 1) Point-to-point links with electrical regeneration (70’)
• 2) Point-to-point links with multiplexing - WDM (80’)
• 3) Point-to-point WDM links with optical amplification (90’)
Rev PA1
TCOM
Rev PA1 8/37
TCOM
Recent advances on transmission
• QAM modulation– 1.568 Tbit/s with a 16-QAM [1]
– Spectral efficiencies > 6 b/s/Hz
• More channels – 500 channels (25 Ghz channel spacing) [2]
• Faster modulations– 640 Gbit/s [3]
[1] Xiang Liu, Sethumadhavan Chandrasekhar, Benyuan Zhu, Peter Winzer, David Peckham “7 x 224-Gb/s WDM Transmission of Reduced-Guard-Interval CO-OFDM with 16-QAM Subcarrier Modulation on a 50-GHz Grid over 2000 km of ULAF and Five ROADM Passes“, ECOC 2010[2] Chun-Ting Lin, “400-Channel 25-GHz-spacing SOI-based planar waveguide demultiplexer employing a concave grating across Cand L-bands”, Optics Express, Vol. 18, Issue 6[3] Hao Hu, Evarist Palushani, Michael Galili, Hans Christian Hansen Mulvad, Anders Clausen, Leif Katsuo Oxenløwe, and Palle Jeppesen, “640 Gbit/s and 1.28 Tbit/s polarisation insensitive all optical wavelength conversion”, Optics Express, Vol. 18, Issue 10
Rev PA1
TCOM
Rev PA1 9/37
TCOM
Optical switching and optical regeneration
• With these througputs, (< 10 Tbit/s) we have pleanty of leeway at the transmission level
• What are the next steps ?
• Optical switching– Available
• MEMS• Filters• Other
• Optical regeneration– Not available (yet?)
Image sources : http://www.fiberopticsonline.com/article.mvc/Alcatels-new-OXC-leverages-bubble-technology-0001https://www.ntt-review.jp/archive/ntttechnical.php?contents=ntr200710sp5.html
Rev PA1
TCOM
Rev PA1 10/37
TCOM
Optical switching• Per channel (lightpath) switching – Optical Circuit Switching (OCS)
After demultiplexing, deflect a wavelength in a particular direction
+ The easiest way
+ Can be done manually and statically
- Low granularity
- More generally… low flexibility
• Sub channel switching – Optical Packet/Burst Switching (OPS/OBS)After demultiplexing, deflect a wavelength in a particular direction for a given
duration
+ Finer granularity, more flexibility
- Network Control overhead
- Switching time overhead
- More complex network Management
Rev PA1
TCOM
Rev PA1 11/37
TCOM
Optical networks applications
• Intra-office communications• Fibre to the home• Application specific (e.g. CERN)
• Long-haul networks– Range up to 5’000 km or more
– Traffic demands ~ 100 Gbit/s per node pair
– Intra domain communications no more than ~100 nodes
– High Availability (six nines 99.9999% 30 sec per year)• One second of unavailability at 1 Tbit/s = 1 Terabit of lost…
– High investments
Not in the scope of this talk
Rev PA1
TCOM
Rev PA1 12/37
TCOM
Optical networks – goals and challenges
GOALS :
• Above all: deliver bits– High throughput– High availability– Low congestions– No errors
• Minimize the costs– Energy– Investments– Maintenance
• Minimize the risks– Network must be under control– Capacity must be available
CHALLENGES :
• Improve network utilisation– Reduce the cascaded
overprovisionning– Minimize the frequency « trap »
• Reduce energy consumption– Avoid per bit operations– Rationalise the utilisation
• Improve network management schemes
– Without adding too much complexity– While guaranteeing availability
Rev PA1
TCOM
Rev PA1 13/37
TCOM
2. Research in optical networks• Components
– Fibres
– Lasers/modulations
– Multiplexers/demultiplexers
– Amplifiers
– Cross-connects
• Network design and planning– Resources provisioning
• Network design
– Resources allocation• Network planning
• Network orchestration– Remote operation
– Computer aided management
– Configuration automation
+
Rev PA1
TCOM
Rev PA1 14/37
TCOM
Routing and Wavelength Assignment - RWA
Most famous optical network problem
Network design version
• Given: – a matrix of demands (# lightpaths required
between each source-destination pair)
• To find: – Network capacities
– a route and a wavelength for each lightpath
• Constraints:– A lightpath must keep the same wavelength
all the way long
– Two lightpaths cannot share a fibre if they use the same wavelength
• Objective:– Minimise the required wavelength
– Minimise the number of required fibres
Network planning version
• Given: – a matrix of demands
– A list of capacities (fibres and λ)
• To find: – a route and a wavelength for each lightpath
• Objective:– Minimise the rejected demands
– Minimise the “frequency traps”
– Minimise the required wavelength
– inimise the number of required fibres
Rev PA1
TCOM
Rev PA1 15/37
TCOM
RWA – more variants
• In the network planning version, the demands may– Arrive
• Simultaneously• At different time points
– Begin• Immediately• Later in time (delayed)
– Last• Forever (open end)• For a fixed duration
• Design + Simultaneous + Immediate + Forever colouring problem• Planning + Simultaneous + Immediate + Forever bin packing [4]• Planning + Simultaneous + Delayed + Fixed duration scheduling [5]
[4] N. Skorin-Kapov, “Routing and wavelength assignment in optical networks using bin packing based algorithms[5] X. Liu, C. Qiao, et al. “Task Scheduling and Lightpath Establishment in Optical Grids
Rev PA1
TCOM
Rev PA1 16/37
TCOM
RWA – event more variants
• Impairments constrained RWA– RWA that takes into account the signal quality and inter wavelength
perturbations [6]
• Protection aware RWA– Both main and spare path must be found– Dedicated or shared protection
• Multi priority RWA
• Multicast aware RWA
• Etc.
[6] A. Marsden, A. Maruta, K.-I. Kitayama, “Routing and Wavelength Assignment Encompassing FWM in WDM Lightpath Networks
Rev PA1
TCOM
Rev PA1 17/37
TCOM
RWA – Solving methods
• ILP / MILP• Constraint Programming• Heuristics
– ILP Relaxation
– Tabu search
– Greedy
– Simulated annealing
– ….
Rev PA1
TCOM
Rev PA1 18/37
TCOM
Routing and Regenerator placements
• Optical signal must be regenerated after a certain journey• Regenerators must be installed
– However, regenerators require more power, more space, more complex chassis, more maintenance
– They should be placed only in strategic places
Problem : minimise the network's critical length (if not fixed)
minimise the number of equipped nodes
minimise the number of regenerations
minimise the routing costs
a
b
c d
ef
g
a-c-b
a-c b-c
a-c-d b-c-d c-d
a-e b-c-a-e c-a-e d-f-e
a-e-f b-c-d-f c-d-f d-f e-f
a-c-g b-c-g c-g d-g e-f-d-g f-d-g
CL=1
CL=2
CL=3In this example, two solutions to reduce CL:
A) Add a regenerator in d
B) Reroute e-f-d-g e-a-c-g
Rev PA1
TCOM
Rev PA1 19/37
TCOM
Problem – Conflicting Objectives
• Minimize the routing shortest paths taking no detours More regeneration sites
• Minimize the # of regenerations shortest path… in most cases
• Minimize the sites oblige lightpath to take detours More regenerations and more routing costs
S
a b
x
c d e
1010 10
88 8
8S
a b
x
c d e
1010 10
88 8
8
(a) (b)
S
a x
y
z
8
8
8
5
5
5
5
5
5
10
10
8
8
8
5
5
5
5
5
5
10
10
S
x
y
z
(a) (b)
a d
c f
b eb
c
d
f
e
(a) Regenerations : 3Routing cost : 3x(8+5+5) = 54
(b) Regenerations : 5Routing cost : 2x(8+5+10) + 18 =
64
Rev PA1
TCOM
Rev PA1 20/37
TCOM
Multi-objective optimisation
0 0.25 0.5 0.750
5
10
15
20
25
30
30 n
odes
C1
pena
lty (
in %
)
0 0.25 0.5 0.75
0
5
10
15
20
25
Fo
50 n
odes
C1
pena
lty (
in %
)
0 0.25 0.5 0.750
10
20
30
40
C2
pena
lty (
in %
)
0 0.25 0.5 0.750
10
20
30
40
Fo
C2
pena
lty (
in %
)
0 0.25 0.5 0.750
5
10
15
20
25
30
# of
reg
. si
tes
0 0.25 0.5 0.750
10
20
30
40
Fo
# of
reg
. si
tes
CP-RegRoutingSites
CP-RegSitesRouting
CP-Sites
CP-SitesRegRouting
CP-SitesRoutingReg
Enum-RegSites
Enum-Sites
CP-RoutingRegSites
CP-RoutingSitesReg
CP-Sites
CP-SitesRegRouting
CP-SitesRoutingReg
Enum-RoutingSites
Enum-Sites
CP-RegRoutingSites
CP-RegSitesRouting
CP-RoutingRegSites
CP-RoutingSitesReg
Enum-Sites
GreedyRegRouting
GreedyRoutingReg
Ro
utin
g p
en
alty
(in
%)
Relative optical reach
Rev PA1
TCOM
Rev PA1 21/37
TCOM
Optical Burst Switching
• General problem of Optical Circuit Switching (of circuits in general) :– A customer of a long-haul operator requires capacity for a fixed duration
– He measures demand peeks of 60 Gbit/s
… whereas the average demand is ~5Gbit/s He nevertheless orders a 30 Gbit/s connection
Overprovisioning factor : 600%
– The operator has many customers
– In general, 25% of them ask for connection
… but sometimes 75% of them The operator is required to design its network accordingly
Overprovisionning factor : 300%
Most of the time, a sixth of a third of the capacity is used… ~5%
Rev PA1
TCOM
Rev PA1 22/37
TCOM
Optical Burst Switching – Statistical multiplexing
• General idea :– With OCS, the operator cannot look “inside” a circuit and fill the voids
– Idea: • Offer to the customer to carry its small packet directly (e.g. IP packets)• Schedule them himself in the network
– Problems:• IP packets are too small to be inserted independently in the optical network• Reserve a circuit for these packets ? Back to the original problem• Multiplex statistically at network edge ?
– Require huge routers (avoid per bit operations ?)– Require to centralise the entering packets (rationalise utilisation ?)
– Solution:• Aggregate in edge nodes smaller packet until reaching an adequate size• Let the network node cores multiplex these bursts
Rev PA1
TCOM
Rev PA1 23/37
TCOM
OBS – General Assumptions
• Burst of about 1mbit – 0.1 ms at 10Gbit/s• Burst are sent in a cut-though manner
– No per bit operation along the way
• Burst are preceded by a Burst Control Packet (BCP) sent in advance– A dedicated lower bit rate channel is reserved for the BCPs
– The BCP “announces” the burst arrival at intermediate nodes
– One way reservation• If a node has no resources, the BCP is dropped, and the burst will be blocked
[7] C. Qiao, M. Yoo, “Optical Burst switching (OBS) – a new paradigm for an optical internet, Journal of high speed networks, 1999 – IOS Press
Rev PA1
TCOM
Rev PA1 24/37
TCOM
OBS Variants
• Burst assembly mechanisms (fixed size, fixed delay, hybrid)• Conventional OBS vs. Emulated OBS
– In E-OBS, BCP and burst are emitted simultaneously• Burst are delayed at each core node entrance
• Reservation protocols– Explicit setup - Explicit release
– Estimated setup - Estimated release
• Scheduling algorithms– With or without void filling
Rev PA1
TCOM
Rev PA1 25/37
TCOM
More OBS Variants
• Routing :– In general, try to minimize the resources consumption shortest path
– Sometimes, better to avoid congested zones• Pro-active routing scheme : load-balancing• Re-active routing scheme : deflection routing
• Synchronous or quasi-synchronous OBS
• Etc.
Rev PA1
TCOM
Rev PA1 26/37
TCOM
QS-OBS : Performance analysis
[8] O. Pedrola, S. Rumley, et al. “Performance overview of the quasi-synchronous operation mode in optical burst switching (OBS) networks, Elsevier Journal of Optical Switching and Networking, Issue 8, In Press
Rev PA1
TCOM
Rev PA1 27/37
TCOM
Contention in OBS
• Contention occurs when a burst cannot be forwarded on its natural path
• Among all the situations causing contention, one can highlight two extreme cases :
Statistical fluctuation – transient phenomenon – short term overload
"bad luck"
No fluctuation! - stationary phenomenon – long term overload
"misconfiguration"
Rev PA1
TCOM
Rev PA1 28/37
TCOM
Contention avoidance methods
Connection Access Control (CAC)
Adaptive load balancing
Traffic engineering – load balancing
Redimensioning
Burst deflection
Flow smoothing and synchronisation
Buffering with FDL
Short term overload
Medium
term overload
Long term overload
Not viable economically
Rev PA1
TCOM
Rev PA1 29/37
TCOM
Our goal
Integrate in one single scheme :
perform this scheme directly at the OBS layer
A Connection Access Control (CAC)
An adaptive load balancing
A burst deflection
Short
Medium
Long
Adaptive burst Admission and Forwarding
[9] S. Rumley, O. Pedrola, et al. “Feedback Based Load Balancing, Deflection Routing and AdmissionControl in OBS Networks”, Journal of Networks, Academy Publisher, Nov 2010
Rev PA1
TCOM
Rev PA1 30/37
TCOM
e1
1
2
3
4
e55
In more detailsID :1-5:ac
ID :1-5:ac1
ID :1-5:ac1,3
- Burst Control Packet (BCP) carries an ID
- BCP contains a list of visited nodes
- When a BCP is dropped…
ID :1-5:ac1,3,4
… or arrives at destination
a feedback is sent to each node of the list
1-5:ac OK1-5:ac OK1-5:ac OK
Rev PA1
TCOM
Rev PA1 31/37
TCOM
In more details
e1
1
2
3
4
e55
- BCP also store the remaining offset time and of course the destination index
- When a core node takes a forwarding decision it adds this decision in a local table (pending table)
- When a core node receives a feedback it retrieves the corresponding decision from the table and updates its feedback table
ID :1-5:acOff: 4Dest: e5 ID :1-5:ac
Off: 3Dest: e5
Next : 3 !
Pending table node 1
3351-5:ac
Next hopOffsetDestID
ID :1-5:acOff: 2Dest: e5
ID :1-5:acOff: 1Dest: e5
ID :1-5:acOff: 0Dest: e5
1-5:ac OK
……………
……………
00335
00212
-+NextOffsetDest
1
Rev PA1
TCOM
Rev PA1 32/37
TCOM
Burst forwarding
e1
1
2
3
4
e55
ID :1-5:acOff: 4Dest: e5
6241335
5134235
-+NextOffsetDest
96.4 %
97.6 %
- When a BCP arrives, the core node retrieve the feedbacks corresponding to the destination and offset
- The success probabilities are estimated for each possible next hop
- Core node tries to make a reservation, starting with the highest probability
- If no reservation is possible on the most favorable next hop, other alternatives are successively tried
ID :1-5:acOff: 3Dest: e5
Rev PA1
TCOM
Rev PA1 33/37
TCOM
Burst admission
e1
1
2
3
4
e55
ID :1-5:acOff: 3Dest: e5
9156425
200325
-+NextOffsetDestIn this case, choosing 3 will lead to burst loss :offset is insufficient (2-3-4-5 3 hops)
ID :1-5:acOff: 2Dest: e5
3 should thus be excluded even there is no other solutionCAC Mechanism :
We assume a threshold TCAC and a minimal number of feedbacks FCAC
If the success estimation E is < TCAC while the number of received feedback is ≥ FCAC next hop is excluded
Rev PA1
TCOM
Rev PA1 34/37
TCOM
3. Alternative research ideas in optical networks
• All-Optical Network Coding– For protection purposes mainly
p1,m p2,m
p1,s
p1+2,s
p1,m p2,m
p1,s p2,s
p1+2,s
[10] E. D. Manley et al. “All Optical Network Coding”, Elsevier Journal of Optical Communications and Networking, Volume 2, Number 4, April 2010
Rev PA1
TCOM
Rev PA1 35/37
TCOM
OBS MultiCast
• Combine (adaptive) deflection routing with Multicast routing?
Rev PA1
TCOM
Rev PA1 36/37
TCOM
Ad-hoc wireless optical networks
• Light beams can also propagate in the air (free space optics)– At relatively high speed on short distances
• Typically 10-40 Gbit/s < 200m• 1-10 Gbit/s < 10km
• Now they have to be manually installed• They might be automated in the future
– Beam tracking
– Ad-hoc topology organisation
• Major problem : almost ON/OFF – Obstacle : OFF
– Not as in Wifi, where obstacle only affect SNR
– Multicast protection required
Image source : http://www.systemsupportsolutions.com/