Last modified 8 years ago Last modified on 11/30/2010 09:08:35 PM

Post-DC3b Schema Changes

LSST Database

Expected table size changes

  • Object:
    • 576 bytes maybe out: covariances might go away: 24 x 6 filters x FLOAT
    • 34 bytes maybe out: IAU
    • at least 80 bytes in: photo redshift: 20 FLOAT for pdf
    • 48 bytes in: Summary statistics for image model residuals, 2 x 6 filterx x FLOAT
    • ~22 bytes in: full classification info (say 2x(BIGINT+SMALLINT)
    • 8 bytes in: petrosian
  • Source:
    • possibly will need to handle multiple versions of calibrated sources
    • possibly will need to handle sources that span multiple ccd exposures (per DataAccWG nov 30, 2010)

Object table

  • Photometric redshift
    • should this be denormalized into separate table and joined with Object?
    • how to represent PDF? Probably ~20 FLOATs will be enough
  • Segmentation info
    • Sloan used: parent, numberOfChildren. We need to represent full graph, and Sloan approach is inefficient.
    • may need to introduce weights
    • may need pointers to postage stamps, especially for deblended objects (blobs in db probably not a good idea)
  • Full classification
    • Beyond star/galaxy, eg. quasars, supernovae etc. Will need to capture weights. Will need separate table.
  • IAU
    • Do we really need it
    • If we do, consider removing classification ("GAL")
  • bounding box
    • rectangle far from ideal, need something more sophisticated
    • now using cannonical (SG) model, need to revisit
  • How are we going to keep track of objects blended with other objects? Related query: dbQuery024?
  • Summary statistics for image model residuals (~2 floats per band)
  • Do we need all the per-filter covariances (requested and used primarily by Jim)
  • summary of various counts - revisit what else would be useful to have beyond uNumObs
  • Revisit if it makes sense to keep ra/decl per model
  • muRaDecl_PS_Cov
  • petrosian...


  • probably none of i* need to be persisted.