june 2012 aa power challenges workshop
TRANSCRIPT
WorkshopJune 2012 AA Power Challenges
WorkshopJune 2012 AA Power Challenges
WorkshopJune 2012 AA Power Challenges
Aperture Arrays
Michael Kramer
WorkshopJune 2012 AA Power Challenges
~ 250 Dishes: 0.3-3GHz
~250 Sparse Aperture Array Stations 70-450MHz
2-Core Central Region
SKA1 construction: 2016-2019
Artist renditions from Swinburne Astronomy Productions
WorkshopJune 2012 AA Power Challenges
~ 250 Dense Aperture Array Stations 400-1450MHz ~ 2700 Dishes: 0.3-10GHz
~250 Sparse Aperture Array Stations 70-450MHz
3-Core Central Region
Artist renditions from Swinburne Astronomy Productions
SKA2 construction: 2020 on
WorkshopJune 2012 AA Power Challenges
Some SKA Experiments…..~ 250 Dense Aperture Array
Stations 400-1450MHz ~ 2700 Dishes: 0.3-10GHz
~250 Sparse Aperture Array Stations 70-450MHz
3-Core Central Region
~ 2700 Dishes: 0.3-10GHz
DishesPrecision pulsar timing• General relativity• Gravity wavesCosmic Magnetic FieldsAstrobiology
WorkshopJune 2012 AA Power Challenges
Some SKA Experiments…..~ 250 Dense Aperture Array
Stations 400-1450MHz ~ 2700 Dishes: 0.3-10GHz
~250 Sparse Aperture Array Stations 70-450MHz
3-Core Central Region
~250 Sparse Aperture Array Stations 70-450MHz
AA‐lowEpoch of Re‐ionisation• Birth of the UniverseHI absorptionTransients
WorkshopJune 2012 AA Power Challenges
Some SKA Experiments…..~ 250 Dense Aperture Array
Stations 400-1450MHz ~ 2700 Dishes: 0.3-10GHz
~250 Sparse Aperture Array Stations 70-450MHz
3-Core Central Region
~ 250 Dense Aperture Array Stations 400-1450MHz
AA‐midBillion galaxy survey ‐ HIPulsar search and timingTransients
WorkshopJune 2012 AA Power Challenges
0
2,000
4,000
6,000
8,000
10,000
12,000
10 100 1,000 10,000 100,000
Sensitivity: Aeff/Tsys m
2 K‐1
Frequency MHz
Sensitivity Comparison
SKA2
SKA1
MeerKAT
LOFAR
ASKAP
eVLA
SKA2
SKA1
LOFAREVLA
Sensitivity Comparison
1
10
100
1,000
10,000
100,000
1,000,000
10,000,000
100,000,000
1,000,000,000
10,000,000,000
100,000,000,000
10 100 1,000 10,000 100,000Survey
Spe
ed : S
ensitivity
2 *FoV A
4 K‐2de
g2
Frequency MHz
SKA2
SKA1
MeerKAT
LOFAR
ASKAP
eVLA
SKA2
SKA1
LOFAR
EVLANote: log scale!
SKA1 & SKA2 will have much higher sensitivity & survey speed than existing instruments
ApertureArrays
Survey Speed Comparison
WorkshopJune 2012 AA Power Challenges
SKA Timeline
2019 Operations SKA1 2024: Operations SKA2
2020‐2024 Construction of Full SKA, SKA2 €1.5B
2016‐2019 10% SKA construction, SKA1 €350M
2012 Site selection
2012 ‐ 2016 Pre‐Construction: 1 yr Detailed design €90MPEP 3 yr Production Readiness
2008 ‐ 2012 System design and refinement of specification
2000 ‐ 2007 Initial concepts stage
1995 ‐ 2000 Preliminary ideas and R&D
WorkshopJune 2012 AA Power Challenges
Type of AA selection
300 500
WorkshopJune 2012 AA Power Challenges
LOFAR station
WorkshopJune 2012 AA Power Challenges
AA-low – Development
WorkshopJune 2012 AA Power Challenges
AA-low outline specification
Parameter SKA1 SKA2 Comments
Type of array Single element Single element Sparse array using a single wide‐band element
No. of elements /station 1750 11,000
No. of elements total 500,000 3,000,000 Approximately
Approx. Size of elements 1x1x2 m 1x1x2 m Must be small enough for the pitch
No. of polarisations 2 2 Each element has two receiver chains
Diameter of station 80m 180m
Number of stations 280 280 Anticipated number SKA Stations
Element communication Analogue fibre Analogue fibre Requires copper for power
Layout pseudo‐random pseudo‐random The most flexible design is as individual elements.
Frequency range 70‐450 MHz 70‐450 MHz May be down to 50MHz
Digitisation rate 1 ‐ 2GS/s 1 ‐ 2GS/s There is no frequency conversion, covers full frequency range with guard bands
Digitisation depth 6 or 8‐bit 6 or 8‐bit Required for RFI environment at these frequencies
Max instantaneous bandwidth
400 MHz 400 MHz Covers operating band of array
Output data rate /station 140Gb/s 8Tb/s Organised as 4+4bit complex data
Data rate into correlator 40Tb/s 2.2Pb/s Peta = 1015
WorkshopJune 2012 AA Power Challenges
AA-low Station
Element
AA-low StationSKA1
......
…....
Cooling
AA‐lowDigitisation &
Station Processing
RFI shielded System clock
Control &Monitoring
StationBeams
PowerGrid
Correlator&
Services
Single or dual fibres
AnalogueFibreSKA1 1,750
Elements
............
Element power distribution
Element power distribution
SKA2 11,000Elements
e/o
Powerconditioning
Elements:50-450MHz
LNA, filter, gain Power over copper
Mixer +
500MHz LO
f500MHz
Pol 2 Pol 1
Data Pol 1 & 2
WorkshopJune 2012 AA Power Challenges
SKA1 AA-low Station Processing
o/e
o/e …
o/e
o/e
RF over Fibrefrom
Elements
ADCs2GS/s6-8 bit
Spectral filters
1stBeam
forming
Station beamform
ing
o/e
o/e …
o/e
o/e
RF over Fibrefrom
Elements
Spectral filters
1stBeam
forming
1750Elements
2 x 70-450MHz
RFI shield
16Gb/sper element
140Gb/s
3 x 56Gb/sInfiniband
e/oData toCentralProcessing System
…
WorkshopJune 2012 AA Power Challenges
Uniboard 1 (Jive)– 16 input signal paths– Input bandwidth: 450 MHz, digitised– Output bandwidth: 40 beams of 300 MHz– 14 layer board; Altera Stratix 4 230: 500 GMAC/s
WorkshopJune 2012 AA Power Challenges
Uniboard 1 Implementation
ADC Interface
Shelf:• 4 Processors• 8 ADC interfaces• 64 inputs
(32 elements)
WorkshopJune 2012 AA Power Challenges
Analogue and Comms PowerProcessing requirement
Processing and digitisation
AA-low SKA1 Station power
Spectral filter:Polyphase filter into 1024 channels 105 MACsPFF rate at 1GS/s 106 /sProcessing rate per element 2*1011 MAC/sTotal spectral filter proc. (1750 el.) 3.5*1014 = 350TMAC/s
Beamforming:Each element 40GS/s (>160Gb/s): 8*1010 MAC/sTotal processing/station (1750 el.): 1.4*1014 = 140TMACs
Total station processing: ~500TMAC/s
Technology FPGA(TMAC/s)
Board (TMAC/s)
# per station*
Power/Board, inc ADC (W) Total
UNIBOARD 1 0.5 4 200 400 80kW
UNIBOARD 2 ~4.0 32 25 500 12.5kW
SKA1 processing 10 est. 80 10 700 7kW*allowance made for inefficiency
Element powerLNA 50mW 100mWGain chain and mux 50mW 100mWOptical Transmission 100mw 150mWTotal Element power 350mWAll elements <1000W
Communications etc. powerTransmission 3*56Gb/s 100WInternal comms 30*56Gb/s 300WMisc. 1000W
Total Station 2.5kW
Total AA-low station power ~10kW 3MW tot
WorkshopJune 2012 AA Power Challenges
Functions:Correlator, beamformer, gridding, imaging, searching, timing & storage
Flexible, upgradeable, development path
Central processing pipelines
…IncomingData fromcollectors
Switch
Buffer store
Switch
Buffer store
HPC BulkStore
CorrelatorBeam
former
UV
Processor
Imaging:
Non‐Imaging:
CornerTurning
CourseDelays
Fine F‐step/Correlation
VisibilitySteering
ObservationBuffer
GriddingVisibilities Imaging
ImageStorage
CornerTurning
CourseDelays
Beamforming/De‐dispersion
BeamSteering
ObservationBuffer
Time‐seriesSearching
Searchanalysis
Object/timingStorage
WorkshopJune 2012 AA Power Challenges
Processing rack and blade
Host processorMulti‐core X86
GPU
‐1 ‐>10TFLOP/s
GPU
‐2‐>10TFLO
P/s
To rackswitches
Disk 1≥1TB
56Gb/s
PCI Bus
Disk 2≥1TB
Disk 3≥1TB
Disk 4≥1TB
Processing blade 1Processing blade 2Processing blade 3Processing blade 4Processing blade 5Processing blade 6Processing blade 7Processing blade 8Processing blade 9Processing blade 10
Processing blade 11Processing blade 12Processing blade 13Processing blade 14Processing blade 15Processing blade 16Processing blade 17Processing blade 18Processing blade 19Processing blade 20
Leaf Switch‐1 56Gb/sLeaf Switch‐2 56Gb/s
42U Rack
Processing Blade:
MIC?
• 20 TFlop• 2x56 Gb/s comms• 4 TB storage• <1kW power
• Capable host (dual Xeon)
• Programmable• Significant RAM
Blade Specification
21
WorkshopJune 2012 AA Power Challenges
Central Processing System
1280 Blades incl. correlator
…
…
…
…
…
…
…
…
…
…
…
…
AA-lowData 1
1
280
AA-lowData 2
1
280
DishesData 4
1 2 16
… 1 3 N
…
HPC
BulkStore
2
Switch
Correlator/UV processor
Further UV processors
Imaging Processor
Corner Turnerswitches
56Gb/s each
…
…
AA-lowData 3
1
280
1
250
……
16 switches36 x 56Gb/s ports
WorkshopJune 2012 AA Power Challenges
Central processing power
Racks Power kW Comments/rack Total
Corner turning switches 2 7 14 Houses both AA and Dish input switches
Correlator/UV processor 48 31 1488 Correlator with UV processing capacity
UV processor per 64GB/s 16 31 496 Number of UV processors to support 64GB/s of visibilities
Central switch 1 10 10 Occupies 29U, but expect rest will be filled with something
Imaging processor per PFLOP/s 5 80 400 Assumes BG/Q, more processing scales linearly
Bulk store with 4.3PB 3 20 60 1.5PB/rack expansion
Total for Basic system 75 2,468 kW
Includes AA-low & dishes
WorkshopJune 2012 AA Power Challenges
SKA Phase 2……
Challenges:• Very high data rates• Large numbers of elements• Substantial processing requirements
100MW !
WorkshopJune 2012 AA Power Challenges
SKA2 wide area data flow
16 Tb/s 4 Pb/s
24 Tb/s
20 Gb/s
20 Gb/s
WorkshopJune 2012 AA Power Challenges
Elements Bunker
AA-low signal path
LNA
Antenna
GainBlock
Analogconditioning
AnalogueSignal Transport
ADC
ADC – Processor comms
Digitisation P
rocessing
Tile – station processoroptical comms
Prim
ary Station P
rocessing
Secondary Station Processing
Wide areaoptical comms
To Correlator
Digitisation Station Processing
optical interconnect
Clock Distribution
e/o o/e
WorkshopJune 2012 AA Power Challenges
SKA2 AA Station processor
PrimaryStation
ProcessorBoard
0
…..
PrimaryStation
ProcessorBoard
1
…..
PrimaryStation
ProcessorBoard
(max 35)
…..
…..
SecondaryStation
ProcessorBoard
0
SecondaryStation
ProcessorBoard
1
SecondaryStation
ProcessorBoard
(max 35)
…..
…..
…..
…..
…..
…..
…..
012
35
ToElement
Digitisation
Long distance drivers
…..
012
Long distance drivers
…..
Long distance drivers
…..
Optical linksTo Correlator
Each link is 12 fibre lanes@10Gb/s
“All to All”Connections
Each link is 12 fibre lanes@10Gb/s
…..
…..
…..
Station Processor
o/eo/eo/eo/eo/eo/e
o/eo/eo/eo/eo/eo/e
o/eo/eo/eo/eo/eo/e
o/eo/eo/eo/eo/eo/e
o/eo/eo/eo/eo/eo/e
o/eo/eo/eo/eo/eo/e
e/oe/oe/oe/oe/oe/o
e/oe/oe/oe/oe/oe/o
e/oe/oe/oe/oe/oe/o
e/oe/oe/oe/oe/oe/o
e/oe/oe/oe/oe/oe/o
e/oe/oe/oe/oe/oe/o
PChip
PChip
PChip
PChip
PChip
PChip
PChip
PChip
PChip
PChip
PChip
PChip
012345
303132333435
Each link is 12 fibre lanes@10Gb/s
To Element digitisationor Primary
Station Processors
To Secondary Station Processors or long distance fibre drivers
Each link is 12 fibre lanes@10Gb/s
Each link is 12 diff. copper lanes@10Gb/s
12-channel Rx module. e.g Avago
AFBR-820BXXZ
12-channel Tx module. e.g Avago AFBR-810BXXZ
Total Raw input data rate: 4.32Tb/s
Total Raw output data rate: 4.32Tb/s max
“All to All”Connections
ControlProcessor
LineTx/R
x
Station Control
Requirements:• High bandwidth in • High bandwidth out• Largely cross connected• Scaleable at various levels• Programmable beamforming
PChip20
TMAC
6 x
120G
b/s
6 x
120G
b/s25 watt
WorkshopJune 2012 AA Power Challenges
SKA2AA-lowPower
Subsystem SKA2 RemarksAntenna element: mW
1 LNA 50 Projection from SKADS work2 Antenna gain block 40
Digitisation and processor:3 Analogue sig conditioning 40 Estimate4 ADC 75 1.0 GS/s each channel5 Clock distribution 10 estimate (less with more ADC per chip)6 Coms: ADC to processor Phase 2: integrated ADC and processor7 Digitiser Processor 300 SKA2: 40 W for 128 receiver inputs8 Digitiser Control ccts etc. 140 SKA2: 25 W for 128 receiver inputs9 Copper comms: Processor to optical driver 10 1.2 W for 128 receiver inputs
10 Optical Coms: Tile to Station proc. 60 SKA1: 4.4 W. SKA2: 2.5 W for 40 receiver inputsStation Processor:
11 Primary Station processor 75 SKA2: 600 W. for 7776 receiver inputs12 Copper comms:
Processor to optical driver10 1.2 W for 128 receiver inputs
13 Optical Coms:Primary to Secondary Station proc.
20 SKA2: 2.5 W for 40 receiver inputs
14 Secondary Station processor 75 SKA2: 600 W. for 7776 receiver inputs15 Copper comms: Processor to optical driver 10 16 Tb/s station output with
153,000 receivers @ 10mW/Gb/sLong Distance comms:
16 Wide area comms Accounted for separately
Electrical power used 915Electrical power supplied @ 85% efficiency 1,076Power incl cooling at 25% cooling power 1,345
Total station power 31 kW
WorkshopJune 2012 AA Power Challenges
AA-mid Array – SKA2
WorkshopJune 2012 AA Power Challenges
Possible AA-mid construction
Beamformer
......
......
......
......
Beamformer
The join!
Membrane
Ground plane
Guideframe
Tile supportGround
Top View
......
......Non-conductingGuideframe
WorkshopJune 2012 AA Power Challenges
Vivaldi
AA-mid elements
FLOTT: (a)(d)
BECA: (b)(e)
ORA: (c)(f)
WorkshopJune 2012 AA Power Challenges
AA-mid proposed signal path
LNA
Antenna
GainBlock
AnalogCond.
Local Tile Analogue Signal Transport
ADC
ADC – Processor comms
Tile Digital Processing
Tile – station processoroptical comms
Primary Station Processing
Secondary Station Processing
Wide areaoptical comms
To Correlator
Tile Processing Station ProcessingFront-end
optical interconnect
Clock Distribution
RF B
eamform
ing
Tile
Tile
Tile
TTD 4-element clusters
Tile processingunit
Element & gain. Phantom power
WorkshopJune 2012 AA Power Challenges
AA-mid designParameter Value Comments
Type of array Single element Dense array using Vivaldi or ORA.
Number of elements 110,000
Pitch of elements 15 cm λ/2 at 1000MHz
No. of polarisations 2 Each element has two receiver chains
Diameter of station 56m
Cluster size 4 elements Uses true time delay beamforming
Tile size 16 x 16 elements Built out of 4 x 4 clusters
No. of Tiles 430 Each tile is ~2.4m square
Number of stations SKA2 250 Anticipated number of Phase 2 SKA Stations
Element communication Copper Includes Phantom power
Layout Dense rectangular Regularly spaced
Frequency range 400‐1450 MHz
Digitisation rate 3GSamples/s There is no frequency conversion,
Digitisation depth 6/8‐bit Required for RFI environment at these frequencies
Beamforming technology Digital Using cluster outputs
Max inst. bandwidth 1000 MHz Covers operating band of array
Max output data rate 16Tb/s Organised as 4+4bit complex data
WorkshopJune 2012 AA Power Challenges
AA-midPower
Subsystem SKA2 RemarksFront‐end (Quads of 4 elements): mW
1 LNA total 120 4xLNAs at 30mW each2 Antenna gain block 120 4 units at 30mW each
Tile beamformer:3 Analogue sig conditioning 160 4 channels at 40mW each4 Analogue beamforming 100 Combiner for 4 elements for digitisation5 ADC 50 2.5 GS/s, 10Gb/s each channel6 Clock distribution 10 estimate (less with more ADC per chip)7 Coms: ADC to processor Integrated ADC and processor8 Tile Processor 200 25 W for 128 digital channels9 Tile Control ccts etc. 100 13 W for 128 digital channels
10 Copper comms: Processor to optical driver 10 1.2 W for 128 digital channels11 Optical Coms: Tile to Station proc. 20 2.5 W for 128 digital channels
Station Processor:12 Primary Station processor 130 600 W for 4608 digital channels13 Copper comms:
Processor to optical driver10 1.2 W for 128 digital channels
14 Optical Coms:Primary to Secondary Station proc.
20 2.5 W for 128 digital channels
15 Secondary Station processor 130 600 W for 4608 digital channels16 Copper comms: Processor to optical driver 10 16 Tb/s station output with
153,000 receivers @ 10mW/Gb/sLong Distance comms:
17 Wide area comms Accounted for separately
Electrical power used 1,190 Per digital channel Electrical power supplied @ 85% efficiency 1,400Power incl cooling at 25% cooling power 1,750
Total station power 96 kW
WorkshopJune 2012 AA Power Challenges
2018+ Technology Shopping listTsys for AA-mid <40K
Analogue system power 100mW per Rx
>3GS/s ADC power <100mW
DSP processor chip performance >20TMACs
DSP Power for 20TMAC ~25W
DSP chip comms, I/O count >128 x 11Gb/s
DSP-digitiser integration Possible
50m optical links, pluggable >120Gb/s
50m 120Gb/s link power 2.5W
Flash storage module capacity 20TB
Many-core streaming processor >50TFlop
50TFlop processor power <300W
Super computer performance >10PFlop
Super computer power <1MW/PFlop
AA Power challenges
can be met with realistic
developments*
* e.g. International Technology Roadmap for semiconductors, ITRS