LOFAR project
description
Transcript of LOFAR project
LOFAR project
Astroparticle Physics workshop
26 April 2004
LOFAR concept
• Combine advances in enabling IT:• inexpensive environmental sensors 10.000’s of sensors • wide area optical broadband networks custom+GigaPort/Géant • high performance computing IBM BlueGene/L
to make a ‘shared aperture multi-telescope’
but also:
to sense and interpret the environment in innovative ways
System spec driver
LOFAR SensorsSensor type ApplicationsHF-antenna: astrophysics
astro-particle physics
VHF-antenna: cosmology, early Universesolar effects on Earth, space weather .
Geophones: ground subsidence .gas/oil extraction
Weather: micro-climate predictionprecision agriculturewind energy .
Water: precision agriculturehabitat managementpublic safety
Infra-sound: atmospheric turbulencemeteors, explosions, sonic booms
LOFARPhase 1
- Radio telescope - Seismic imager - Precision weather for agriculture, wind energy
Sensor field
Central processor
Fibre data transport
Integrate LOFAR networkinto regional fibre network,sharing costs with schools,
health centres etc.
Radio Telescope Specifications
• Frequency range: – 20 – 80 MHz, 120 – 240 MHz
• Angular resolution– few – 10 arcsec
• Sensitivity– 100x previous instruments at these frequencies
• Shared aperture multi-telescope– up to 8 independent telescopes
• plus geophone, weather etc arrays– operated from remote Science Operations Centers
• similar to LHC ‘tier-1’ centers
One day in the life ofLOFAR, the radio
telescope
Tel
esco
pe n
r.
Challenges
• Data rate• ~ 15 Tbits / sec total data generated (increasing later)
• ~ 330 Gbits / sec input data rate to central processor
• ~ 1 Gbit / sec to distributed Science Operations Centres
• Computational resources• ~ 34 TFLOP/s in custom co-processor (IBM BG/L)
• ~ 500 TBytes on-line temporary storage
• Calibration• adaptive multi-patch all-sky phase correction
• 10 sec duty cycle
VC Beam formerCalibration Beam s
VirtualCoreSubStation beam s(full bandwidth)
Rem oteStation beam s
phase,gainparameters
(time dependent)
SC-Selfcal
T EM PORARYDAT A
ST ORE
Station-Core Visibilities
Synchronisation &route
Synchronisation &route
C entra l P rocessor : D ataFlow P rocess ing C entra l P rocessor : S torage C entra l P rocessor : D ataS et A na lys is
pre-process
FringeCorrection
Calibration
Imaging
Data Archive
Instrum entM odel
EnvironmentM odel
Sky M odel
QuickView
Central RFIMitigation
Station-Core Correlator
Station-Station Correlator
CorrectedStation-StationVisibilities
Store:25 Gbps
Store:25 Gbps
Input rate> 300 Gbps
Input rate> 300 Gbps
Products
~1 Gbps
Products
~1 Gbps
3 T-ops
5 T-flops
2 T-ops
Transpose~300 Gbps
Storage: >500 TB
Within correlator:20 Tbps
15 T-ops
IBM BlueGene/L
• IBM– 1st research machine on road to multi-peta-FLOP/s
– 3 BG/L machines under construction LLNL, LOFAR, IBM Research
– numbers 1-10 of Top-500 supercomputers in one machine (LLNL)
– SOC technology, standard components for reliability– dual PowerPC 440 chips per node with 700 MHz clock
– scalability– to many times 100.000 nodes
– low power, air cooled– ~ 20W per node
IBM BlueGene/L
• LOFAR– BG/L is our 1st non-custom central processor
• total CPU power is ‘interesting’ (34 TFLOP/s) and scalable• component failure rate: one every 3 months, DRAM dominated
– BG/L is embedded co-processor in LINUX cluster– stripped down LINUX kernal on-chip– general purpose capability allows complex modelling on-line, real time
– efficient for complex arithmetic, streaming applications• 330 Gb/s input data rate initially; 768 Gb/s max
– low power• 150 kW for LOFAR ( 6k nodes )
– scalable beyond LOFAR to SKA requirements
Tier-0 computing LHC, LOFARin 2006
CPU(SPECint95)
No. of Processors
Disk storage (TB)
Tape storage (PB)
LAN throughput
(Gb/s)
LHC / exp’t x 4 exp’ts( Tier-0 )
2,8 106
5600 / 11200
(?)
2160
12
368
LOFAR( EOC )
3,4 106
6144 / 12288
~ 500
??
> 330
LOFAR with Bsik financing
Central core- plus -
45 stations150 km max baseline
Mid-LOFARwould extend into
Lower Saxony,Schleswig-Holstein,
Northrhein-Westphalen
Max-LOFARwould have stations from Cambridge UK
to Potsdam DE,from Nançay FR
to Växjö SE
1-10 Gbps
China
USA
South Africa
Russia
Post-2005:JIVE + LOFAR data processing centre
30 Gbps – 2 Tbps
LOFAR, the Sensor Network is under consideration as FP7
‘Technology Platform’
LOFAR project timeline
• PDR in June/Oct 2003: M€ 14 expended• Dutch funding end 2003: M€ 52 for ‘infrastructure’
• funding must be matched by ‘partners’ – 18 member consortium: additional partners possible
• formal goal is economic positioning w.r.t. ‘adaptive sensor networks’– RF, seismic, infra-sound, wind-energy sensors
• prototyping of a full station is in progress• 100 low frequency antennas in field, now are making all-sky videos• end 2004, expect 2 beam web-based system on-line (to gain experience)
– issues: calibration, RFI, adaptive re-allocation of resources• BlueGene/L delivery in 1Q-2005
• FDR start in mid-2004, complete mid-2005• procurement start mid-2004, end mid-2006
• Initial operational status: end-2006 (solar minimum)• full operational status: mid-2008
Remaining tasksfor which partners are being sought
• Array configuration size: new stations !– extension of array size to 400+ km is highly desirable
• cost is ~ € 500k per station• fiber connections through Géant, national academic networks
• Definition, designation of operations centers– Science Operations Centers are remote, on-line
• basic data taking and archiving of observations• financing mostly local, plus contribution to common services
– Engineering Operations Center in Dwingeloo• monitor system, perform maintenance• integrated operations team (with WSRT, possibly JIVE)
• Operational modelling and User interface• use of (quasi-real-time) GRID technologies foreseen• work packages not funded / manned yet
Where?
Where?
User involvement
• Test User Group• Heino Falcke, leader
– Lars Bähren, Michiel Breintjens, Stefan Wijholds etc
• ‘open’, ‘remote’ access to developing system– step-wise functionality improvements until 2006
• 1st user workshop Dwingeloo, May 24-25, 2004• ASTRON is ready to host a (limited) number of
young researchers to test, help develop the system
• Formal operations from 2007• scheduling will be an ‘interesting’ problem
LOFAR Research Consortium
Universities Research Institutes Commercial
Univ. of Amsterdam
ASTRON(management org.)
Ordina TechnicalAutomation bv
TU Delft CWI Dutch Space bv
TU Eindhoven IMAG Twente Institute for Wireless and Mobile Communications bv
Univ. of Groningen KNMI Science[&]Technology bv
Leiden Univ. TNO-NITG
Nijmegen Univ. LOPES Consortium
Uppsala Univ. MPIfR-Bonn