The Square Kilometre Array & High speed Astrophysics Group, Cavendish Laboratory, University of Cambridge http://www.mrao.cam.ac.uk/ bn204/ CRISP Annual Meeting FORTH, Crete, Greece March 2012
Outline
SKA Dishes Artist impression of SKA dish receptors Credit: SPDO/Swinburne Astronomy Productions
Sparse Aperture Array Artist impression of SKA aperture array receptors Credit: SPDO/Swinburne Astronomy Productions
Dense Aperture Array (SKA 2 ) Artist impression of SKA dense aperture array receptors Credit: SPDO/Swinburne Astronomy Productions
Inner core configuration... Many in the central 5 km core Credit: SPDO/Swinburne Astronomy Productions
... at centre of a continent-scale But also many very long baselines Site decision expected in next few months!
In summary: 100 better sensitivity than current best 10 6 sky survey speed then current facilities Frequency coverage 70 MHz 10 GHz Extremely radio quite site either in Western Australia or Karoo Desert in South Africa Staged construction: Pathfinders: ASKAP/MeerKAT under construction SKA1 : Construction start in 2016 SKA 2 : Construction start in 2019 Will require Exascale computing to form images and analyse them
SKA Key Science Drivers Origins Neutral Hydrogen from the Epoch of Re-Ionisation to now First stars and galaxies Galaxy evolution Role of Active Galactic Nuclei Dark Energy, Dark matter Cradle of Life Fundamental Forces Pulsars, General Relativity & Gravitational waves Origin & evolution of cosmic magnetism
Example SKA science galaxy evolution to z 10 Hubble VLA 3000 galaxies 15 radio sources
Example SKA science galaxy evolution to z 10 Hubble SKA simulation 3000 galaxies
Outline
Radio Interferometers Point source S Incoming radiation wavefront Digitised timeseries data Digitised timeseries data Digitised timeseries data Digitised timeseries data X X X X X X X X Correlator X X X X X X X X Raw UV Data
Visibilities and sky images Measure the product (visibility) of signals from each pair of (baseline): V ( B ) = E1 E 2 = I( ( ) B S S ) exp iω (1) c Ignoring polarisation: dldm V ij = 1 l 2 m A ij(l, m)i(l, m) { [ 2 ( )]} exp i2π ul + vm + w 1 l 2 m 2 1 (2) For small field of view, approximately: V ij dldm A ij (l, m)i(l, m) exp {i2π [ul + vm]}, (3) i.e., there is a Fourier relationship between the sky image and measured visibilities.
Main stages Correlation of signals from antennas to form visibilities N 2 problem, large data flow rates Gridding of irregularly sampled uv-data onto a cubic grid Memory locality problem, challenge to parallelise Formation of sky-images from uv cubes Need to achieve for dynamic range of 10 7, deconvolution Calibration, i.e., determination of A ij (l, m) coefficients, which must be done from the astronomical images Iterative algorithms
Outline
Wide area dataflow for SKA 2
Example analysis of post-correlator data rates SKA 1 Aperture Array EoR Experiment 44GBs 1 RFI Calibration Correlator Gridder Imager 0.5GBs 1 no RFI 70GB/major cycle 114GB/obs Science 114GB/obs Image postprocessing / analysis Raw UV data UV cubes Spectral cubes
Example analysis of post-correlator data rates SKA 1 Aperture Array EoR Experiment Calibration loop 44GBs 1 UV raw data store RFI Calibration Correlator Gridder Imager 0.5GBs 1 no RFI 70GB/major cycle 114GB/obs Science 114GB/obs Image postprocessing / analysis Raw UV data UV cubes Spectral cubes
UV Data Store Summary Holds the raw uv (i.e., ungridded) data for duration of each observation max duration abut 5 hours Enables iterative algorithms on the raw uv data calibration Multiple iterations over the full data set required Read performance must be substantially faster then write All interfacing systems fully observatory controlled few data protection/security issues
UV Data Store Requirements I Write throughput: Maximum expected write data rate of 330 GBs 1 Read throughput: Maximum expected read data rate of 1650 GBs 1 Storage duration: Expected duration of storage of data is 5 hours Storage capacity: Required capacity is about 6 PB Hardware interfaces: Industry standard hardware interfaces suitable for use with supercomputers Data format Most likely custom binary format. Possibly HDF5 or some derived data format Security No security/quota/user permission required as both ends directly controlled by SKA software
UV Data Store Requirements II Real-time: Soft real-time write operation. Consistent, predictable performance for bot reads and writes required. Reliability: Loss of binary data not a problem as long as correctly flagged and does not cause hold-ups to processing Timeframe: Full production beginning 2017. Design reviews in 2015.