simulating proto-clusters utilizing the austrian grid
DESCRIPTION
Michaela Lechner Eelco van Kampen Daniel Clarke Simon Ostermann Manchester, May 2007. Simulating Proto-clusters utilizing the Austrian Grid. Institute of Astro- and Particle Physics Distributed and Parallel Systems Group Institute for Computer Science. - PowerPoint PPT PresentationTRANSCRIPT
Simulating Proto-clusters utilizing Simulating Proto-clusters utilizing the Austrian Gridthe Austrian Grid
Michaela LechnerMichaela Lechner Eelco van Eelco van KampenKampen
Daniel ClarkeDaniel Clarke
Simon OstermannSimon Ostermann
Manchester, May Manchester, May 20072007
Institute of Astro- and Particle Institute of Astro- and Particle PhysicsPhysics
Distributed and Parallel Distributed and Parallel Systems GroupSystems Group
Institute for Computer ScienceInstitute for Computer Science
Protocluster Science Protocluster Science is Sub-mm Scienceis Sub-mm Science
Sub-mm:Sub-mm:200200μμm - m - 1mm 1mm
MillimeteMillimeter:r:1mm - 1mm - 10mm10mm
Sub-mm needed forSub-mm needed forHigh-z galaxy formationHigh-z galaxy formationHigh-z clusters (structure High-z clusters (structure formation?)formation?)
Technology just starting to Technology just starting to mature, breakthrough mature, breakthrough inevitable.inevitable.JCMT citation rate rivals JCMT citation rate rivals HST!HST!
At 850At 850m, a galaxy has m, a galaxy has same flux density from same flux density from zz = = 1 - 101 - 10
James Clerke Maxwell TelescopeJames Clerke Maxwell Telescope
Largest existing sub-millimeter one dish Largest existing sub-millimeter one dish telescope in the world (diameter of 15m)telescope in the world (diameter of 15m)
The JCMT is used to study our Solar The JCMT is used to study our Solar System, interstellar dust and gas, and System, interstellar dust and gas, and distant galaxies. distant galaxies.
Situated close to the summit of Mauna Situated close to the summit of Mauna Kea, Hawaii, at an altitude of 4092m (high Kea, Hawaii, at an altitude of 4092m (high & dry). & dry).
SCUBA: The Submillimetre Common-SCUBA: The Submillimetre Common-User Bolometer Array User Bolometer Array
JCMT
SHADES: SCUBA Half Degree SHADES: SCUBA Half Degree Extragalactic SurveyExtragalactic Survey
Lockman Hole East Subaru/XMM-Newton Deep FieldLockman Hole East Subaru/XMM-Newton Deep Field
It is not yet possible to measure the clustering properties of sub-It is not yet possible to measure the clustering properties of sub-mm sources. Redshift determination is currently in progress. mm sources. Redshift determination is currently in progress. ((radio correlation)radio correlation)
More sub-mm in the FutureMore sub-mm in the Future
ALMA: Atacama Large Millimeter ArrayALMA: Atacama Large Millimeter Array
Herschel Space ObservatoryHerschel Space Observatory
SCUBA-2 (starting January 2008)SCUBA-2 (starting January 2008)
HSOHSO
ALMAALMA
complete: 2012complete: 2012
operating: 2009operating: 2009launched: 2007launched: 2007
SCUBA-2 on JCMTSCUBA-2 on JCMT
large 8 x 8 arcmin field-of-viewlarge 8 x 8 arcmin field-of-view
Simultaneous imaging at 450 and Simultaneous imaging at 450 and 850μm850μm
Fully-sampled images of the sky in Fully-sampled images of the sky in <4 seconds <4 seconds
bring “CCD-style” imaging to the bring “CCD-style” imaging to the JCMT for the first timeJCMT for the first time
20, 40 and 100 arc surveys are 20, 40 and 100 arc surveys are currently being planned using currently being planned using SCUBA-2 building on the SCUBA-2 building on the understanding developed with understanding developed with SHADES and SCUBA1SHADES and SCUBA1
Simulating ObservationsSimulating Observations
CDM model of van Kampen, CDM model of van Kampen, Rimes & Peacock (2004)Rimes & Peacock (2004)
Ingredients for a Ingredients for a Semi-numerical Semi-numerical Galaxy FormationGalaxy Formation::
Cosmological model Cosmological model (standard)(standard)
Halo formation and Halo formation and merger historymerger history
Gas dynamics and Gas dynamics and radiative coolingradiative cooling
Star formation and Star formation and stellar feedbackstellar feedback
Stellar population Stellar population synthesissynthesis
Accurate Dust modelAccurate Dust model
Simulating ObservationsSimulating Observations
only only onceonce
very very fastfast
GRASILGRASIL
Initial Initial data set data set currently currently created created on HPC on HPC ClusterCluster
Ingredients for a Ingredients for a Semi-Semi-numerical Galaxy numerical Galaxy FormationFormation::
Cosmological model Cosmological model (standard)(standard)
Halo formation and Halo formation and merger historymerger history
Gas dynamics and Gas dynamics and radiative coolingradiative cooling
Star formation and Star formation and stellar feedbackstellar feedback
Stellar population Stellar population synthesissynthesis
Accurate Dust modelAccurate Dust model
The Dust Model: GRASILThe Dust Model: GRASIL
GraGraphite and phite and SilSilicone: icone: principal contributors to principal contributors to galactic dustgalactic dust
Considers the physical effects Considers the physical effects of graphite, silicate and PAH of graphite, silicate and PAH particles on a galaxy’s SEDparticles on a galaxy’s SED
by Laura by Laura SilSilva (Gian Luigi va (Gian Luigi GraGranato) 1999, FORTRAN 95nato) 1999, FORTRAN 95Calculates the complex line Calculates the complex line integral between volume integral between volume elements as a photon travels elements as a photon travels through the galaxy. through the galaxy. Computationally intensive! Computationally intensive! Small code, small data filesSmall code, small data files
Outputfiles: SED Data (1-2 Outputfiles: SED Data (1-2 kB)kB)Problem Problem ideally suited for ideally suited for Grid Computing!Grid Computing!
BulgeBulgeMolecular cloudMolecular cloud
DiskDisk
Diffuse ISM, Free stars and CirrusDiffuse ISM, Free stars and Cirrus
Equatorial planeEquatorial plane
Small code, small data filesSmall code, small data filesOutputfiles: SED Data (1-2 Outputfiles: SED Data (1-2 kB)kB)All galaxies independent from All galaxies independent from each othereach other
Simulating ObservationsSimulating Observations
Creation of Galaxy Formation model Creation of Galaxy Formation model predictions:predictions:
Comparing predictions with future Comparing predictions with future observationsobservations
Current lightcone with 60 timeslices à ~ Current lightcone with 60 timeslices à ~ 20.000 galaxies 20.000 galaxies -> -> 1 million galaxies1 million galaxies
Predicting whole lightcone Predicting whole lightcone -> -> usable for all usable for all telescopes and wavelength regimestelescopes and wavelength regimes
Protoclusters:Protoclusters:
Only interested in part of the lightcone Only interested in part of the lightcone (relevant timeslices)(relevant timeslices)
Looking into the pastLooking into the pastP. Heinämäki, I. Suhhonenko, E. Saar, M. Einasto, J. Einasto, and H. Virtanen
Time
Time
Time
Time
SPITZERSPITZER
HerschelHerschel
ALMAALMA
SCUBA 2SCUBA 2
including noise including noise and observational and observational effectseffects
With a survey With a survey field of 100 field of 100 square degrees, it square degrees, it seems seems statistically likelystatistically likely, , based on dark based on dark matter matter simulations simulations that that both an over both an over density region density region and blank field and blank field will be observed will be observed early in the early in the survey.survey.
. .
Mock SCUBA-2 survey: ‘super-SHADES’ with Mock SCUBA-2 survey: ‘super-SHADES’ with 100 square degrees100 square degrees
This is This is what we what we currently currently havehave
SCUBA-2 legacy surveysSCUBA-2 legacy surveys
2 years
5 years
Simulating ObservationsSimulating Observations
ΛCDM Model
Galaxy Formation
Model
Cosmological Parameters
(fixed)
Galaxy Formation Parameters (free)
ΛCDM Files
Galaxy Properties
Galaxy Properties
Galaxy Properties
GRASIL Parameters (free)
ASKALON
GRASIL GRASIL GRASIL
Dust Adapted SED
Dust Adapted SED
Dust Adapted SED
Observation Formation
Instrument and Filter Parameters
Austrian Grid
Porting to the Austrian GridPorting to the Austrian Grid
Splitting one big parameter file into small Splitting one big parameter file into small parameter files for each galaxy parameter files for each galaxy
in generation of initial data set in generation of initial data set
in GRASILin GRASIL
Compiling GRASIL in 3 different flavors:Compiling GRASIL in 3 different flavors:generic 32 bitgeneric 32 bit
AMD64AMD64
Itanium 2Itanium 2
Porting to the Austrian Grid reduces computational Porting to the Austrian Grid reduces computational time per galaxy to approximately 4.2 seconds. time per galaxy to approximately 4.2 seconds. (Lechner et al. 2007)(Lechner et al. 2007)
ASKALON WorkflowASKALON Workflow
Simple workflow: Simple workflow:
Input Data Input Data Streamer ActivityStreamer Activity
Parallel LoopParallel Loop
Collector ActivityCollector Activity
GRASIL Activity GRASIL Activity deployed on several deployed on several Grid-sites, Grid-sites, ‘embarrassingly parallel’.‘embarrassingly parallel’.
ASKALON WorkflowASKALON Workflow
Preparation of data: Ruby script creating Preparation of data: Ruby script creating jobset-tarballs jobset-tarballs (future ASKALON will do it (future ASKALON will do it automatically)automatically)
less overheadless overhead
more balancingmore balancing
(list of galaxies is not continuous)(list of galaxies is not continuous)
Auto deployment of GRASILAuto deployment of GRASIL
Performance monitoring/predictionsPerformance monitoring/predictions
Increasing code efficiencyIncreasing code efficiency
Calculating the correlation Calculating the correlation matrix of the input matrix of the input parameters suggests that parameters suggests that the following parameters the following parameters are co-correlated with are co-correlated with execution time:execution time:
Host machine Host machine Radius of the Disk Radius of the Disk componentcomponentCold gas massCold gas massCold gas metallicity Cold gas metallicity Total stellar massTotal stellar massBulge densityBulge density
Testrun with 25000 Galaxies on Testrun with 25000 Galaxies on different Grid sitesdifferent Grid sitesFor each host type, there is a clear For each host type, there is a clear clustering of effective execution time. clustering of effective execution time.
ASKALONASKALONLocal Installation or as a Java WebserviceLocal Installation or as a Java Webservice
Autodeployment in ASKALONAutodeployment in ASKALON
Blank FieldBlank Field
Inserted proto-cluster @ z=2.5Inserted proto-cluster @ z=2.5
A proto-cluster at A proto-cluster at z≈2.5 significantly z≈2.5 significantly boosts the number boosts the number of visible 850 of visible 850 μμm m sources, compared sources, compared to a field-only map. to a field-only map. Both maps are half a Both maps are half a square degree in square degree in size with a size with a resolution according resolution according to the JCMT beam.to the JCMT beam.
Matching sub-mm galaxy number Matching sub-mm galaxy number countscounts
Mock with no (proto)-cluster
Mock with 1 (proto-)cluster
Mock with 2 (proto-)clusters
Underlying figure from Coppin et al. (2006)
Finding proto-clusters in the Finding proto-clusters in the SCUBA-2 and Herschel surveysSCUBA-2 and Herschel surveys
Number overdensity Number overdensity of sub-mm sources of sub-mm sources for a field containing for a field containing a rich proto-cluster a rich proto-cluster increases with the increases with the flux cut.flux cut.
Conclusions IConclusions I
Sub-mm observations is probably one of the most Sub-mm observations is probably one of the most important wavebands in modern cosmologyimportant wavebands in modern cosmologyWith the next generation of Sub-mm surveys it With the next generation of Sub-mm surveys it seems likely that observations of over density seems likely that observations of over density regions will become commonregions will become commonAccurate simulations of proto-clusters will help us to Accurate simulations of proto-clusters will help us to understand the physics behind cluster and galaxy understand the physics behind cluster and galaxy formation.formation.For simulations to ‘keep pace’ with detector For simulations to ‘keep pace’ with detector technology developments, new computing technology developments, new computing techniques need to be adapted.techniques need to be adapted.
UrgencyUrgency: Comparison with observations soon!: Comparison with observations soon!
Extreme case of Extreme case of parallelizationparallelization, huge amount of , huge amount of calculation power needed.calculation power needed.
Input for Grid Middleware (ASKALON) improvementInput for Grid Middleware (ASKALON) improvement
Conclusions IIConclusions II
Clustering is detected in SHADES, but with Clustering is detected in SHADES, but with large uncertaintieslarge uncertainties
Redshifts or large surveys are needed to Redshifts or large surveys are needed to improve upon this: the SCUBA-2 and/or improve upon this: the SCUBA-2 and/or Herschel legacy survey(s)Herschel legacy survey(s)
Bright sub-mm sources are mostly Bright sub-mm sources are mostly associated with high-density regions (proto-associated with high-density regions (proto-clusters and the like)clusters and the like)
the contribution of (proto)-clusters to sub-the contribution of (proto)-clusters to sub-mm source countsmm source counts
‘ ‘shallow’ (> 12 mJy) surveys are sufficient shallow’ (> 12 mJy) surveys are sufficient to find high-z (proto-)clustersto find high-z (proto-)clusters
Thank you!Thank you!