Click to add title

Download Report

Transcript Click to add title

40 Gb/s and 100 Gb/s
Technologies for
Research & Education Networks
Tom McDermott
Fujitsu
July 17, 2007
All Rights Reserved, ©2007 Fujitsu Network Communications
Agenda

Growing Bandwidth Requirements


Motivation for 40 Gb/s
Motivation for 100 Gb/s

Client Interfaces and Network Interfaces
 Characteristics of optical transmission systems
 Technical approaches for modulation
 Some experiments
1
All Rights Reserved, ©2007 Fujitsu Network Communications
LHC Data Collection

LHC currently plans on 10GE clear optical wavelengths for
world-wide grid-computing network.
 When LHC comes on-line, it is predicted to produce 12 to 14
petabytes / year*
14 petabytes per year 






~ 11.2 * 1016 bits /  * 107 seconds
~ 3.6 * 109 bits / second
Keeps one 10GE busy pretty much full-time forever…
But real data collection is usually very bursty in time
40G & 100G would offer numerous practical benefits
• Data set exchange and backup, Load sharing, etc.
*HP News release, Jan 27, 2004
2
All Rights Reserved, ©2007 Fujitsu Network Communications
Motivation for 40 Gb/s

Support 40G Packet-Over-SONET (POS) router interfaces.


Avoiding inverse-multiplexing or link-aggregation issues.
System Capacity Expansion: 4 x 10GE, 4 x OC-192





Existing DWDM Systems support ~ 40 channels in C-band, with 10
Gb/s per wavelength.
Typical system has 100 GHz (0.8 nm) spacing between optical
channels.
Filter bandwidths ~ 65 gHz ( ±0.325 gHz) at half-dB down points
40 Gb/s can ‘fit’ within existing DWDM channel filters and ROADMs if
the modulation bandwidth can be constrained.
Permits 4x capacity expansion without additional fibers or terminals.
3
All Rights Reserved, ©2007 Fujitsu Network Communications
Motivation for 100 Gigabit Ethernet

Today: Link Aggregation (802.3ag) approach for N x 10 GE.
 Must not re-order packets within any specific flow.
 How to identify flows, and force each flow to just one of the
multiple 10GE lanes?
Today: Equal Cost Multi Path -- A way to (almost) randomly assign
each flow to one of the 10GE lanes.
 Dynamically it breaks: all the big flows fight for the same one lane.
Performance degrades to 10GE for milliseconds to seconds.
 Need a better solution.

Tomorrow: Use native 100 GE link. No need to ‘hash’ flows.
 IEEE 802.3 HSSG: 4-5 approaches being considered





802.3: Low cost, high volume, short range (300m) are key drivers
10-lane byte-striped, 1-lane x 10 colors, 5 lanes, even copper is being
proposed.
Low cost ~10 kM also being considered.
Question: how to transport 100GE on DWDM systems?
4
All Rights Reserved, ©2007 Fujitsu Network Communications
IEEE HSSG Status

Call for Interest (CFI) for Higher Speed Signalling Group
[HSSG] approved by IEEE plenary in 2006.
 Charter is to define an IEEE standards project:


Technical need and economic justification of a new standard.
Successful CFI results in an IEEE Project Authorization
Request (PAR). Time frame roughly 2 years to a standard
(unless it gets derailed). Estimate ~4Q09.


Potential detours / derailments: 40GE, 40+kM
Short Range Ethernet connections that are cost-effective and
producible.




Typically 300-1000 meters. Connecting Host to Network, Data center.
IEEE objective: 10x the speed at 3x the cost.
Current proposals: Copper, 4x CWDM, 5x VCSEL parallel, 10x parallel,
serial.
Also typically specifies a medium-range interface ~ 10 kM.
5

Current Proposals: 1x (serial), 2x, 4x, 5x parallel.
All Rights Reserved, ©2007 Fujitsu Network Communications
Client & Network Interfaces
Service-provider’s network
NE
Ethernet
Service
Switch
NE
NE
NE
Client Interface – IEEE 802.3
Proposed Standards Activity:
• Inter-vendor
interoperable
• Lower performance,
shorter distance
• Higher volume
• Lower cost
• Link status
Network Interface – Not a proposed
standard:
• High performance, longer
distance
• Minimize wavelength
consumption
• Manageable, FEC counts
6
All Rights Reserved, ©2007 Fujitsu Network Communications
100 Gb/s Network Interface options

Utilize multiple wavelengths, for example 10 x 10 Gb/s.






Requires delay equalization across the 10 wavelengths due to
differential propagation delay.
Consumes ¼ of the C-band.
Need to manage a band of wavelengths as a single service
Lower cost T/R module, but extremely expensive in terms of fiber
utilization.
May be difficult for carrier to justify in a commercial network.
Utilize single wavelength




Single optical wavelength must carry 100 Gb/s of information
Complex T/R, but extremely efficient in terms of fiber utilization.
Easier for carrier to manage as a single wavelength.
If the T/R cost is reasonable, much easier to justify in a commercial
network.
7
All Rights Reserved, ©2007 Fujitsu Network Communications
100 Gigabit Ethernet: How?

Today: DWDM systems designed for 100 GHz channel spacing.
Channel passband: roughly 65 GHz.
Cascading ROADMs narrows the channel from there.
Traditional amplitude modulation methods have upper and
lower sidebands.
Limits data rate to roughly 30 Gb/s per channel.
 New approach needed
Higher-order modulation: 2 bits / symbol
Polarization multiplexing: 2 symbols at the same time.
Can theoretically get us to 100~120 Gb/s per 100 GHz channel.
50 Ghz channel spacing theoretically handles 50~60 Gb/s per
channel (4PSK).
 Narrow channels not as well suited for 100 GE







8
All Rights Reserved, ©2007 Fujitsu Network Communications
Channel Filtering
~65 GHz
Mag
Single ROADM passband
Cascade of ROADM passbands
through multiple nodes
Traveling through
multiple nodes reduces
equivalent filter width
l
ROADM passband
9
100 Gbit/s NRZ signal
25 Gbit/s NRZ signal
200 GHz
50 GHz
100 GigaSymbol/s Modulation:
Spectrum too wide to fit filter
25 GigaSymbol/s Modulation:
Spectral width fits OK
All Rights Reserved, ©2007 Fujitsu Network Communications
Reducing the Symbol Rate

Utilize multiple bits / symbol constellation




Utilize orthogonal optical polarizations



QPSK, 8PSK, 16QAM, etc.
Common technique for radio & wireline, where S/N can be very high.
But optical devices are noisy and the channel is nonlinear, so it’s
difficult to expand the constellation very much.
Vertical and horizontal polarizations do not crosstalk too much.
Transmit two independent signal sets at the same time.
Combination of the two



QPSK: 2 bits / symbol
Polarization Multiplexing: 2 symbols at the same time
Yields 25 GigaSymbols per second on each channel.
• Fits within ROADM filter bandwidths
• Electronics is capable of operating at 25 GigaSymbol/s rate.
10
All Rights Reserved, ©2007 Fujitsu Network Communications
100 Gigabit Ethernet: Transmission

At 100 Gsym/s: transmission impairments are severe.
 Example: Chromatic Dispersion:


100 Gsym/s is 100 times worse than 10 Gb/s.
25 Gsym/sec is 6.25 times worse than 10 Gb/s.

Approach: Coherently demodulate the optical signal:
In-phase and Quadrature-phase components (I and Q).
In essence a radio receiver with an optical front-end.
 Many Optical distortions then linearly map to baseband
distortions.
 Use baseband DSP to un-do some of these effects.
11
All Rights Reserved, ©2007 Fujitsu Network Communications
Coherent Baseband Equalization
(OFC 2007 OTuA1)

Optical signal is mixed with local oscillator


Producing In-phase (I) and Quadrature-phase (Q) signals, for
Both vertical and horizontal polarization: VI, VQ, HI, HQ.

Digitize all 4 signals with high-speed AD converter.
 Baseband processor utilizes adaptive digital filters to remove
optical distortions.
 VLBA telescope array uses this conversion technique at radio
frequencies for mm-wave radio astronomy.
12
All Rights Reserved, ©2007 Fujitsu Network Communications
Key Technologies

Key Technologies for 100GE:





Advantages to this approach:




Multi-level modulation.
Polarization Multiplexing.
Coherent Receiver.
DSP at baseband.
Compatibility with existing transmission systems
Ease of adding 100 GE to existing services without disruption
Economic use of facilities for 100GE
Disadvantages:

More complex T/R module than parallel optical lanes.
13
All Rights Reserved, ©2007 Fujitsu Network Communications
14
All Rights Reserved, ©2007 Fujitsu Network Communications