Pan-STARRS1 Transient Database K. Smith, D. R. Young, S. Valenti, S. Smartt Wednesday, 27 October 2010 PanSTARRS1 Overview • 1.8m telescope • 1.4 Gigapixel Camera • 7 square deg FOV • Image Processing Pipeline (IPP) • Few Tb of data every night Wednesday, 27 October 2010 Survey Modes : Medium Deep Wednesday, 27 October 2010 Survey Modes : Medium Deep • 10 x 7 squ. deg. fields • Observed in g,r,i,z,Y every 4 nights • 113 (g,r) or 240 (i,z,Y) sec exposures • Well known fields • 8 exposures/field daily • 4 or 5 fields observed daily Wednesday, 27 October 2010 Survey Modes : Medium Deep QUB MD Transient Processing Stack–Stack Differencing • Nightly stacks of 8MD exposures per MD field produced by IPP • Differenced with a Reference Stack by IPP •(may be another nightly stack or deep stack) • Download & process 19MB/day StackStack Difference catalogues • (33, 40, now 55 columns of info per FITS table + header) • 80,000 difference objects/day • MD Transient Database grows by 27MB/day • Download & store 200MB/day “postage stamp” images • Daily download of IPP (GPC1) database (4GB compressed) Wednesday, 27 October 2010 Current Medium Deep Coverage Wednesday, 27 October 2010 Processing PS1 Data Wednesday, 27 October 2010 Processing PS1 Data Wednesday, 27 October 2010 Download Catalogue data Processing PS1 Data Wednesday, 27 October 2010 Download Catalogue data Stack-Stack diff catalogues Processing PS1 Data Wednesday, 27 October 2010 Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Stack-Stack diff catalogues Processing PS1 Data Wednesday, 27 October 2010 Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Promote and Assign Object Name or Discard Object Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Promote and Assign Object Name or Discard Object Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Promote and Assign Object Name or Discard Object Crossmatch Promoted Objects (e.g. with CfA) Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Promote and Assign Object Name or Discard Object Crossmatch Promoted Objects (e.g. with CfA) Wednesday, 27 October 2010 Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Automated Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Promote and Assign Object Name or Discard Object Automated Wednesday, 27 October 2010 Crossmatch Promoted Objects (e.g. with CfA) Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Processing PS1 Data Download Catalogue data Pre-Ingest Cuts, Run Crossmatch & Classifier Algorithm, Ingest Data Post-Ingest Quality Cuts & Post Quality Candidates to Eyeball Automated Generate Lightcurve & Scatter Plots for Quality Candidates Request Images for Quality Candidates Poll for & download images Human eyeballing (+ Galaxy Zoo) Manual Promote and Assign Object Name or Discard Object Automated Wednesday, 27 October 2010 Crossmatch Promoted Objects (e.g. with CfA) Stack-Stack diff catalogues Supernova, AGN, NT, Variable Star, Orphan quality detections = 00000010010110 observation filters = yzzgrirgiizgry Software Components Wednesday, 27 October 2010 Software Components CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components MySQL Database CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components HTTP PS1 Difference Catalogues (FITS) PS1 Image Processing Pipeline MySQL Database CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components PS1 Image Processing Pipeline HTTP CCFits CFITSIO PS1 Difference Catalogues (FITS) MySQL Database CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components PS1 Image Processing Pipeline HTTP CCFits HTM MySQL++ CFITSIO PS1 Difference Catalogues (FITS) MySQL API MySQL Database CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components PS1 Image Processing Pipeline HTTP Command Line/CRON PERL Wrapper / File SpliLer C++ Ingester / Classifier CCFits HTM MySQL++ CFITSIO PS1 Difference Catalogues (FITS) MySQL API MySQL Database CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components PS1 Image Processing Pipeline HTTP Command Line/CRON PERL Wrapper / File SpliLer Python PS1 Image Request/Fetch C++ Ingester / Classifier CCFits HTM MySQL++ MySQLDB Python API MySQL API MySQL Database CFITSIO PS1 Difference Catalogues (FITS) CentOS 5 (RHEL 5) Wednesday, 27 October 2010 Software Components PS1 Image Processing Pipeline HTTP PS1 Postage Stamp Server HTTP Command Line/CRON PERL Wrapper / File SpliLer Python PS1 Image Request/Fetch C++ Ingester / Classifier CCFits HTM MySQL++ MySQLDB Python API MySQL API MySQL Database CFITSIO PS1 Difference Catalogues (FITS) CentOS 5 (RHEL 5) Wednesday, 27 October 2010 PS1 Image Stamps Software Components PS1 Image Processing Pipeline HTTP PS1 Postage Stamp Server Command Line/CRON Web Interface PERL Wrapper / File SpliLer Django Framework C++ Ingester / Classifier Apache Python PS1 Image mod_python Request/Fetch CCFits HTM MySQL++ MySQLDB Python API MySQL API MySQL Database CFITSIO PS1 Difference Catalogues (FITS) CentOS 5 (RHEL 5) Wednesday, 27 October 2010 HTTP PS1 Image Stamps Hardware Transient Server & Local IPP DELL PowerEdge 2950 III (64GB) 8 Core Xeon X5460 (3.16 GHz) 1 TB (RAID 5) Internal SAS Storage Transient Server Database Logs & Temporary Tables Web & File Server DELL PowerEdge 2950 III (2GB) 4 Core Xeon E5405 (2.0 GHz) DELL PowerVault MD3000 5.3 TB (RAID 5) SAS Storage Array Transient Server Database Tables DELL PowerVault MD1000 11.2 TB (RAID 5) SATA Storage Array Database Backup Staging Image Stamps Follow-up Data Wednesday, 27 October 2010 DELL PowerVault MD1000 11.2 TB (RAID 5) SATA Storage Array Backup Mirror IPP Data Products (catalogue files) External Catalogue Ingestion Wednesday, 27 October 2010 External Catalogue Ingestion 8 Generic Catalogues 2 x 109 rows 0.5 TBytes (MyISAM) Veron 12 2MASS XSC SDSS Stars 2MASS PSC Guide Star 2.3 NED SDSS Photo Galaxies SDSS Spectroscopic Galaxies Wednesday, 27 October 2010 External Catalogue Ingestion 8 Generic Catalogues > 20 Medium Deep Field Catalogues 109 2x rows 0.5 TBytes (MyISAM) > 40 views 105 3x rows 0.1 TBytes (MyISAM) MD01 NED Veron 12 MD05 NED MD01 Chiappetti 2005 2MASS XSC MD05 Brunner 2008 MD06 NED MD01 Pierre 2007 SDSS Stars MD02 NED 2MASS PSC Guide Star 2.3 NED SDSS Photo Galaxies MD07 NED MD02 Giacconi 2002 MD07 Laird 2009 MD02 LeFevre 2004 MD07 Nandra 2005 MD02 Leher 2005 MD08 NED MD02 Virani 2006 SDSS Spectroscopic Galaxies MD08 Manners 2003 MD03 NED MD09 NED MD04 NED MD04 Hasinger 2007 MD04 Trump 2007 Wednesday, 27 October 2010 MD10 NED External Catalogue Ingestion 8 Generic Catalogues > 20 Medium Deep Field Catalogues 109 2x rows 0.5 TBytes (MyISAM) > 40 views 105 3x rows 0.1 TBytes (MyISAM) MD01 NED Veron 12 MD05 NED MD01 Chiappetti 2005 2MASS XSC MD05 Brunner 2008 MD06 NED MD01 Pierre 2007 SDSS Stars MD02 NED 2MASS PSC Guide Star 2.3 NED SDSS Photo Galaxies MD07 NED MD02 Giacconi 2002 MD07 Laird 2009 MD02 LeFevre 2004 MD07 Nandra 2005 MD02 Leher 2005 MD08 NED MD02 Virani 2006 SDSS Spectroscopic Galaxies MD08 Manners 2003 MD03 NED MD09 NED MD04 NED MD10 NED MD04 Hasinger 2007 MD04 Trump 2007 • Source data in various formats (CSV, FITS,VO Tables) • Custom ingest scripts written (PERL, C++, Java) • Policy adopted of NaN, ±∞, null values = NULL • Unit sphere cartesian coords (cx, cy, cz) calculated on ingest for cone searches • HTM IDs for each data point calculated on ingest Wednesday, 27 October 2010 Local MySQL Databases Crossmatch Catalogues IPP GPC1 Ingest Database (views onto Crossmatch Database) Access to GPC1 Database Django Web Database (views onto Ingest Database) Wednesday, 27 October 2010 '( )) %& %& '*%+, '*%+, 2 !%& 3(- '%& - '( )) # ! *%+, (( )) (#- (./++ ##- %& Transient Server Database Schema '*%+, '*%+, #2 ) !(' #'( ) '%& '( ) # ! *%+, '(# 0(' '(# ) '((1-2 ##- (( )) ##- %& (',/, !('./++ ) !%& ./++ ( " (#- ! " ./++ (#- (./++ !%& # 4 # $ %& !%& !%& ! " Wednesday, 27 October 2010 !%& $% &*$+, &*$+, $% &$% " '' ()) ""- Transient Server Database Schema "2 ) "&' ) &' ) *$+, ""- $% &' ()) *$+, &' ()) $% &*$+, &*$+, '& &$% " 0'& &''1-2 ) '' ()) ""- &'" &'" ) '"- ! ./++ '"-'./++ $% $% '&,/, '&./++ $% " 3 " # Wednesday, 27 October 2010 Crossmatching: JHU Hierarchical Triangular Mesh Splits the Celestial Sphere into “Spherical Triangles” which are spatially numerically adjacent e.g. Level 20 HTM id for RA, DEC = 243.334412243, 53.9783926374 N112212112030032100002 (1474714251366610) circleRegion API returns a vector of HTM ID tuples (high and low HTM ID) Wednesday, 27 October 2010 HTM + Cartesian Cone Search Wednesday, 27 October 2010 HTM + Cartesian Cone Search Wednesday, 27 October 2010 HTM “circleRegion” + Cone Search SELECT * FROM catalogue WHERE ( htm_id BETWEEN id(0) AND id(1) OR htm_id BETWEEN id(2) AND id(3) OR . . OR htm_id BETWEEN id(n-1) AND id(n) ) AND ( cx * object x coordinate + cy * object y coordinate + cz * object z coordinate >= cos(radius in radians) ); Wednesday, 27 October 2010 MySQL Database Organisation MyISAM storage engine utilised (Fast SELECT, fast INSERT, fast replication) (*updates must be carefully managed - table-level locking) RA/DEC and cx, cy, cz Coordinate columns indexed HTM ID column indexed HTM IDs spatially numerically adjacent, table ordered by HTM ID (ALTER TABLE catalogue ORDER BY htmID) Database tables stored on high speed, low latency disks (SAS) (Indexes will eventually be stored on physically separate volumes) Wednesday, 27 October 2010 Contextual Information: e.g. MD08 Wednesday, 27 October 2010 Contextual Information: e.g. MD08 Wednesday, 27 October 2010 Contextual Information: e.g. MD08 Wednesday, 27 October 2010 Contextual Information: e.g. MD08 Wednesday, 27 October 2010 Contextual Information: e.g. MD08 Wednesday, 27 October 2010 Contextual Information: e.g. MD08 Wednesday, 27 October 2010 PS1 Catalogue Ingester - Algorithm Implementation • • Search field algorithms descended from single abstract class “Factory” class determines which algorithm to use on PS1 object catalogue ingest as they are processed Wednesday, 27 October 2010 Wednesday, 27 October 2010 Pre-Ingest Cuts - “Bad” IPP Detection Bitmask SOURCE_BAD_MASK = 1111 0000 0000 0111 1111 1111 1001 1000 PM_SOURCE_MODE_... MSB OFF_CHIP ON_GHOST ON_SPIKE SIZE_SKIPPED X X X X X X X X X BELOW_MOMENTS_SN SKYVAR_FAILURE SKY_FAILURE MOMENTS_FAILURE EXT_LIMIT CR_LIMIT SATURATED DEFECT BADPSF EXTERNAL BLEND SATSTAR X X POOR FAIL X X X LSB Wednesday, 27 October 2010 Post-Ingest Automated Cuts Wednesday, 27 October 2010 Post-Ingest Automated Cuts 1. Set Quality Flag for each object where: XY moments of objects < 1.2 PSF instrumental magnitude error < 0.22 PSF instrumental magnitude error > 0.0 Wednesday, 27 October 2010 Post-Ingest Automated Cuts 1. Set Quality Flag for each object where: XY moments of objects < 1.2 PSF instrumental magnitude error < 0.22 PSF instrumental magnitude error > 0.0 2. Promote objects to Eyeball List where: RMS position scatter < 0.5 arcsec Quality detections >= 3 within 7 observation window Quality detection filters >= 2 Wednesday, 27 October 2010 Post-Ingest Automated Cuts 1. Set Quality Flag for each object where: XY moments of objects < 1.2 PSF instrumental magnitude error < 0.22 PSF instrumental magnitude error > 0.0 2. Promote objects to Eyeball List where: RMS position scatter < 0.5 arcsec Quality detections >= 3 within 7 observation window Quality detection filters >= 2 quality detections = 00000010010110 observation filters = yzzgrirgiizgry Wednesday, 27 October 2010 Post-Ingest Automated Cuts 1. Set Quality Flag for each object where: XY moments of objects < 1.2 PSF instrumental magnitude error < 0.22 PSF instrumental magnitude error > 0.0 2. Promote objects to Eyeball List where: RMS position scatter < 0.5 arcsec Quality detections >= 3 within 7 observation window Quality detection filters >= 2 observation window = 7 quality detections = 00000010010110 observation filters = yzzgrirgiizgry Wednesday, 27 October 2010 Post-Ingest Automated Cuts 1. Set Quality Flag for each object where: XY moments of objects < 1.2 PSF instrumental magnitude error < 0.22 PSF instrumental magnitude error > 0.0 2. Promote objects to Eyeball List where: RMS position scatter < 0.5 arcsec Quality detections >= 3 within 7 observation window Quality detection filters >= 2 observation window = 7 quality detections = 00000010010110 observation filters = yzzgrirgiizgry quality detections in 3 filters Wednesday, 27 October 2010 Python Web Interface Template candidate.html <tr> {% for column in table.columns %} <th> {% if column.sortable %} <a href="?sort={{ column.name_toggled }}"> {{ column }} </a> {% if column.is_ordered_reverse %} <img src="/ps1/site_media/images/up.jpg" /> {% else %} <img src="/ps1/site_media/images/down.jpg" /> {% endif %} {% else %} {{ column }} {% endif %} </th> {% endfor %} </tr> View views.py def candidate(request, tcs_transient_objects_id): transient = get_object_or_404(TcsTransientObjects, pk=tcs_transient_objects_id) initial_queryset = WebViewRecurrentObjectsPresentation.objects.filter(transient_object_id = transient.id) table = WebViewRecurrentObjectsPresentationTable(initial_queryset, order_by=request.GET.get('sort', 'mjd_obs')) return render_to_response('psdb/candidate.html',{'transient' : transient, 'table' : table}) Model models.py class WebViewRecurrentObjectsPresentation(models.Model): id = models.IntegerField(primary_key=True, db_column='id') transient_object_id = models.IntegerField(db_column='transient_object_id') mjd_obs = models.FloatField(db_column='mjd_obs') RA = models.FloatField(db_column='ra_psf') DEC = models.FloatField(db_column='dec_psf') mag = models.FloatField(db_column='mag') filter = models.CharField(max_length=90, db_column='filter') flags = models.CharField(max_length=90, db_column='flags') cmf_file = models.CharField(max_length=90, db_column='cmf_file') image = models.CharField(max_length=90, db_column='filename') class Meta: db_table = u'psdb_web_v_recurrent_objects_presentation' @property def flags_bin(self): a = bin(int(str(self.flags)),32) return a Wednesday, 27 October 2010 Web Interface - Human Cuts Wednesday, 27 October 2010 Wednesday, 27 October 2010 Wednesday, 27 October 2010 Wednesday, 27 October 2010 Wednesday, 27 October 2010 Horror Show Bad Subtraction Kernel Parameters Bad Template Bad Registration - Dipoles Bright Star artifacts Wednesday, 27 October 2010 0D07aak SN Ic, z=0.116 Target Wednesday, 27 October 2010 Reference Difference Crossmatching with CfA Objects Wednesday, 27 October 2010 ~ 100 Spectroscopically Confirmed SNe Wednesday, 27 October 2010 PS1 MD Data Processing - 25/10/2010 PS1 Catalogue Files Processed (3.1GB) Total Objects Processed Rejected Objects (bad flags) Unique Objects ingested Recurrent Objects ingested Total Thread Time (seconds) 39,714 13,299,196 4,020,614 3,941,472 5,337,110 1,398,243 Daily Object Processing Capacity (8 Threads) ~ 6,500,000 Wednesday, 27 October 2010 PS1 MD Data Processing 2010 Unique Objects 3,941,472 Object Type Orphan Variable Star Nuclear Transient AGN Supernova Wednesday, 27 October 2010 Number 2,483,784 1,058,433 195 1,609 397,451 PS1 MD Data Processing 2010 Objects passing automated quality thresholds Unique Objects 26,314 Object Type Orphan Variable Star Nuclear Transient AGN Supernova Wednesday, 27 October 2010 Number 1,366 20,441 6 176 4,218 PS1 MD Data Processing 2010 Objects further refined by human examination (Confirmed/Good/Possible Objects) Unique Objects 1,359 Object Type Orphan Variable Star Nuclear Transient AGN Supernova Wednesday, 27 October 2010 Number 298 255 0 112 694 MD Transient Server Run Time • • • • • • • Download Catalogues (10 minutes) Download GPC1 Database (10 minutes to 3 hours) Backup Ingest DB (45 minutes) Organise files for ingest & Run Classifier (3.5 hours) Run Post Ingest cuts (40 minutes) Request Postage stamps (5 minutes) Download Postage stamps (3-10 hours) Wednesday, 27 October 2010 Outlook for MD Survey • Implement Dipole Stats and other new columns in pre and post-Post Ingest Cuts • Re-examine pre-ingest flag cuts & examine new flags in latest IPP release • Include more catalogues for crossmatching • Convergence with CfA Transient Server Wednesday, 27 October 2010 Survey Modes : 3π Wednesday, 27 October 2010 Survey Modes : 3π +90o • Survey Area = 30,000 sqr. deg. • • • • • avg 1,600 sqr. deg. / night 30 sec exposures ~220 exposures/night 1 - 2 Tb/night o -30 Whole sky 4 times a year in five filters g,r,i,z,Y Wednesday, 27 October 2010 Survey Modes : 3π Wednesday, 27 October 2010 Survey Modes : 3π Difference Catalogue forecast Based on Medium Deep stats so far: • ~ 220 Object Catalogues produced per night in grizY (1.5 GB/day) • ~ 16,000 objects / exposure • ~ 3,500,000 rows per day to process • ~ 9GB of postage stamp image data / day Wednesday, 27 October 2010 Survey Modes : 3π 3π Transient Searching • Difference imaging with static sky – start end of 2010 start of 2011 ? • What can we do before then ? • 3 Faint Galaxy SN survey : • Started in May 2010: • 7 SNe and 4 AGNs Wednesday, 27 October 2010 15 good candidates (spectroscopic follow-up): Survey Modes : 3π 3π Faint Galaxy Supernova Survey • ~ 220 Object Catalogues produced per night in griz •97% “magicked” •35-40% in Sloan areas •200,000 objects per catalogue file (i.e. per exposure) •50,000 after pre-ingest optimisation • No differencing done • Download & crossmatch 6GB/day object catalogues •(33, 40, now 55 columns of info per FITS table x 60 + header) • FGSS database still being developed - no growth stats yet... • Download & store 20MB/day “postage stamp” images Wednesday, 27 October 2010 3π Faint Galaxy Supernova Survey - Download all the 3π-catalogs produced by IPP (smf files) - Select Sloan fields and calibrate the PS1 magnitudes to the Sloan magnitude. - Cross-correlate all the good* sources with the SDSS catalogs and select all the sources for which the closer SDSS object is classified as a galaxy and falls within 3 arcsec of the PS1-3π detection. - Extract all PS1 detections that meet the following criteria: 15 < PS1 mag (observed filter) < 20 19 < sloan mag (r) < 23 distance (sloan galaxy - PS1 detection) < 3 arcsec PS1 mag 2 mag brighter than sloan mag (in the obs. filter) occurrences >= 3 and 2 arcsec grouping of detections Wednesday, 27 October 2010 FGSS SNe 4 Termonuclear SNe 3 Core Collapse (IIL,IIP,II) AGN Wednesday, 27 October 2010 FGSS crossmatching • Cross-match computation done in memory and achieved by dividing the candidates into subsets • (chip by chip) (optimise the cross-match) (140,000 Sloan galaxies per catalogue file) • each spatial cross-match among one catalogue file and the sloan galaxies in the same area selects: • r ~ 524 crossmatches i ~ 870 crossmatches • g ~ 345 crossmatches z ~ 348 crossmatches • ~ 0.7 good detection / each catalogue file (Magnitude and distance cut) • 171 candidates (recurrence cut and grouping of detections) • ~ 10% of 171 selected for observations: – ~ 5% spectroscopically classified as SNe – ~ 2.5 % spectroscopically classified as AGN Wednesday, 27 October 2010 PS1 detections - Sloan catalog • Wednesday, 27 October 2010 Query Sloan chip by chip Running Time • Download files: 9h / day ? ~215 smf files per night ~208 smf files magiked 208 x (140s-180s) / smf file (30.5 M) ~ 8h-10h • Cross-match: 10h / day ? • Zero point - sloan area check ~ 208 x time sloan field check = 30m - 1h30m • (38%) in the sloan area ~ 80 x time cross-match = • <1h> Other steps: (4h - 17h) < 9h > 3h/month ? • Selection cut + web-page (without stamps) 30m • Stamp servers request-download (~ 200 stamps per months) ??? • Eyeballing (50 candidates per months) • Download extra stamps (5-10 candidates per month ~150 stamps) ? Wednesday, 27 October 2010 1h 2-5 good candidates per month Outlook for 3π Survey • • • • • • Fully implement FGSS database • Crossmatching efficiency may need to improve 3π diffs should start to flow in December/January • This may replace the FGSS, or the two may run in parallel for a time Improve pre-ingest cuts - this reduces CPU time in crossmatching algorithms (but at expense of not ingesting the data row) Tune the code to minimise unnecessary latency (e.g. instantiation of unnecessary classes) Liaise with Networks staff about improving data links to server Throw more processors at the problem • Most latency so far is CPU bound Wednesday, 27 October 2010 Wednesday, 27 October 2010