An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M....
-
Upload
juniper-wright -
Category
Documents
-
view
215 -
download
1
Transcript of An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M....
![Page 1: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/1.jpg)
An OptoelectronicNeural Network Packet
Switch Scheduler
K. J. Symington, A. J. Waddie, T. Yasue,M. R. Taghizadeh and J. F. Snowdon.
http://www.optical-computing.co.uk
![Page 2: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/2.jpg)
Outline• Packet switch scheduler.
• Previous demonstrator has proven system feasibility.
• Current demonstrator enhances functionality and performance.
• Motivation.
• Implementation and scalability.
• Conclusions.
![Page 3: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/3.jpg)
The Assignment Problem
Solution is computationally intensive.Neural networks are capable of solving the assignment problem.Their inherent parallelism allows them to outperform any other known method at higher orders.
Can be found in situations such as:• Network service management.• Distributed computer systems.• Work management systems.• General scheduling, control or resource
allocation.
![Page 4: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/4.jpg)
Crossbar Switching
![Page 5: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/5.jpg)
A size N crossbar switch has the same number of inputs as outputs: i.e. m=n=N.
Crossbar Switching
![Page 6: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/6.jpg)
Crossbar Switching
• Packets stored in buffer until output free.
• Packets can request any output line.
• Buffer depth very important.
• Real traffic tends to be ‘bursty’.
![Page 7: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/7.jpg)
Crossbar Switching
• Channel operation exclusive.
• Maximum capacity of N packets per switch cycle.
![Page 8: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/8.jpg)
Crossbar Switching
• Packet can only pass when crosspoint set.
• N2 crosspoint switches required.
• Generic crossbar switch architecture.
![Page 9: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/9.jpg)
Crossbar Switching• Neural network chooses optimal set of packets.
• One neuron required for every crosspoint.
![Page 10: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/10.jpg)
Crossbar Switching
![Page 11: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/11.jpg)
Banyan Switching
![Page 12: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/12.jpg)
• Routing input 2 to output 2 allows only 1 packet to pass.Solution is sub-optimal.
Solution Optimality
24
2
• Routing input 2 to output 2 allows only 1 packet to pass.Solution is sub-optimal.
• Routing input 2 to output 4 and input 4 to output 2 allows 2 packets to pass. Solution is optimal.
![Page 13: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/13.jpg)
The Neuron• Inputs taken from the outputs of
other neurons.
![Page 14: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/14.jpg)
The Neuron• Inputs taken from the outputs of
other neurons.
• Synaptic weights multiply inputs.
![Page 15: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/15.jpg)
The Neuron• Inputs taken from the outputs of
other neurons.
• Synaptic weights multiply inputs.
• Inputs are summed and bias added.
![Page 16: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/16.jpg)
The Neuron• Inputs taken from the outputs of
other neurons.
• Synaptic weights multiply inputs.
• Inputs are summed and bias added.
• Transfer function f(x) performed before output.
![Page 17: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/17.jpg)
Neural Algorithm
2
CyByAt1txtx
n
ikkj
n
jkikijij
ijxijij
e1
1xfy
xij: Summation of all the inputs to the neuron referenced by ij: including the bias.
yij: Output of neuron ij.
A, B and C: Optimisation parameters.
‘Iterations to Convergence’ is an important parameter.
Iterations related to, but not necessarily equal to, time.
: Controls gain of neuron.
Next state defined by:
Neural transfer function:
![Page 18: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/18.jpg)
Neural Interconnect
![Page 19: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/19.jpg)
Convergence Example
Start state – all requested neurons are on.
![Page 20: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/20.jpg)
Convergence Example
1/3 Evolved: Neurons (2, 4) and (4, 2) are beginning to inhibiting neuron (2, 2).
![Page 21: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/21.jpg)
Convergence Example
2/3 Evolved: Neuron (2, 2) is nearly off.
![Page 22: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/22.jpg)
Convergence Example
Fully Evolved. Optimal solution reached.
![Page 23: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/23.jpg)
• Neural network scalability limited in silicon.
• Optoelectronics allows scaleable networks.
• Free-space optics can be used to perform interconnection.
• Only transfer function f(x) need be performed in silicon.
• Input summation is done in an inherently analogue manner.
• Noise added naturally.
Why Optoelectronics?
![Page 24: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/24.jpg)
The VCSEL Array• Optical output element.
• A laser that emits from the surface of the substrate.
• High optical output powers.
![Page 25: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/25.jpg)
The VCSEL Array• Each neuron has one VCSEL for
optical output.
• Performance: Capable of >1GHz operation.
• Scalability: Currently N=16.
![Page 26: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/26.jpg)
Detector Arrays• Optical input element.
• Available in a wide range off the shelf.
• Performance: >1GHz.
• Caveat: faster detectors require more power.
![Page 27: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/27.jpg)
Diffractive Optic Elements (DOEs)
• Large fan-outpossible.
• Efficiency:~50-60%.
• Non-uniformity:<3%.
• Period Size:90µm.
These elements are used as array generators and interconnection elements.
![Page 28: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/28.jpg)
Crossbar switch interconnect.Banyan switch interconnect.
Optical Interconnect
DOE interconnect is space invariant.
![Page 29: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/29.jpg)
VCSEL Array
Lens Ø10mm f=100mm-150mm
DOE Ø15mm Period 90µm
Photodetector Array
Optical System
![Page 30: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/30.jpg)
First Generation System
• Constructed using discrete components.
• Lacked ability to prioritise packets: can lead to channel saturation.
• Uses similar optical system (~330mm).
![Page 31: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/31.jpg)
Current System
• System uses 4×40MHz Texas Instruments 320C5x DSPs.
• DSPs perform transfer function.
• Transfer function fully programmable.
• Reduction of hardware by digital thresholding.
![Page 32: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/32.jpg)
System Scalability
![Page 33: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/33.jpg)
Digital vs. Analogue
Analogue: Optimal ~97%. Digital: Optimal ~91%.
![Page 34: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/34.jpg)
Crossbar Switch ResultsHistogram of packets routed successfully in a crossbar switch.
![Page 35: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/35.jpg)
Banyan Switch ResultsHistogram of packets routed successfully in a banyan switch.
![Page 36: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/36.jpg)
Mean Packet Delay
![Page 37: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/37.jpg)
Mean Packet Delay
• ISLIP4 cannot be implemented larger than N=16.
![Page 38: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/38.jpg)
Mean Packet Delay
• ISLIP4 cannot be implemented larger than N=16.
![Page 39: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/39.jpg)
3 Major effects to consider:• Active effects: <1Hz thermal changes and component
creep.• Static effects: Tolerances in fabricated components
could lead to misalignment in final system.• Adaptive effects: Vibrational effects >1Hz - e.g.
10kHz.Solutions:• Measurement and correction of focusing and positional
error in real time (active optic alignment or adaptive optics).
• Commercially viable: e.g. personal CD player, ASDA £22:95.
• Pre-packaged, pre-aligned modules.
Engineering Issues
![Page 40: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/40.jpg)
Encapsulated System
R. Stone, J. Kim and P. Guilfoyle,“High Performance Shock Hardened Optoelectronic Communications Module”, OC2001, Lake Tahoe,pp. 105-107.
![Page 41: An Optoelectronic Neural Network Packet Switch Scheduler K. J. Symington, A. J. Waddie, T. Yasue, M. R. Taghizadeh and J. F. Snowdon. .](https://reader035.fdocuments.in/reader035/viewer/2022070410/56649f1d5503460f94c34423/html5/thumbnails/41.jpg)
Conclusions• Performance of 100MHz feasible, 1GHz foreseeable.• Scalability mainly limited by VCSEL array size (N=16).• Scalability independent of number of inputs/outputs (N).• A digital system running at 1GHz could supply 2.5 million
switch configurations per second.• Second generation builds on first in that it supports
prioritisation.• What good is a truck without a steering wheel?• Further work:
• Smart pixel implementation and packaging.• Examination of QoS provided by scheduler.• FPGA or custom ASIC implementation using optical interconnects.• Novel neural algorithms and learning.