Atomic access to large arrays

Hi

I'm storing historical meteorological gridded data from GFS (http://www.nco.ncep.noaa.gov/pmb/products/gfs/) into an array field in a table like this:


Dimensions of each value of field vl_grid are (lat:361 x lon:720 = 259920 cells} for a grid of 0.5 degrees (about each 55 Km) around the world. So, vl_grid[y][x] stores the value at dh_date of a meteorological variable specified by id_inventory in the geodesic point


The reverse formula for the closest point in the grid of an arbitary geodesic point will be


Field vl_grid is stored in the TOAST table and has a good compression level. PostgreSql is the only one database that is able to store this huge amount of data in only 34 GB of disk. It's really great system. Queries returning big rectangular areas are very fast, but the target of almost all queries is to get historical series for a geodesic point


In this case, atomic access to just a cell of each one of a only few thousands of rows becomes too slow.

Please, could somebody answer some of these questions?

• It's posible to tune some TOAST parameters to get faster atomic access to large arrays?

• Using "EXTERNAL" strategy for storing TOAST-able columns could solve the problem?

• Atomic access will be faster if table stores vectors for data in the same parallel instead of matrices of global data?
  
  CREATE TABLE grid_f_data_i2 (
    //Specifies the variable and other features of data
    id_inventory integer REFERENCES grid_d_inventory(id_inventory),
    //A new grid is available each 3 hours since 5 years ago
    dh_date timestamp,
    // nu_parallel = y = Round((latitude+90) * 2
    smallint nu_parallel,
    //Values are scaled to be stored as signed integers of 2 bytes
    vl_parallel smallint[],
  CONSTRAINT meteo_f_gfs_tmp PRIMARY KEY
    (co_inventory, nu_parallel, dh_date)
  );
  
  
• There is another faster solution?
  

Thanks in advance and best regards

--
Víctor de Buen Remiro
Tol Development Team member
www.tol-project.org