Grib Collections

A GRIB Collection is a collection of files that have GRIB records in them. Each file will have a grib index (gbx9) file generated for it. Once generated, gbx9 files never have to be redone.

Each logical collection has a CDM index (ncx2) file. All the records for a single runtime comprise a logical collection, called a RC.

case 1. {collection} has single runtime

<dataset name="GRIB Collection Example for tutorial (single runtime)" serviceName="gribServices" >
 <featureCollection name="Ismoro Single Runtime" featureType="GRIB1" harvest="true" path="Grib/srt">
  <collection name="collsrt" spec="/machine/tds/data/grib/ecmwf/.*grib$" />
  <update startup="test" />
 </featureCollection>
</dataset>

Single RC dataset, single CDM index file. One time dimension. May have multiple groups (horizontal domains).

case 2. {collection} has multiple runtimes

<dataset name="GRIB Collection Example for tutorial (multiple runtimes)" serviceName="gribServices">
 <featureCollection name="GFS multiple runtime" featureType="GRIB2" harvest="true" path="Grib/mrt">
  <collection name="collmrt" spec="/machine/tds/data/grib/gfsmrt/.*grib2$" />
  <update startup="test" />
 </featureCollection>
</dataset>

One RC dataset for each runtime, plus PofGC "collection dataset" with both runtime and time dimensions.

case 3. {collection} has multiple runtimes, one per file: use file partition

<dataset name="GRIB Collection Example for tutorial (file partition)" serviceName="gribServices">
 <featureCollection name="GFS file partition" featureType="GRIB2" harvest="true" path="Grib/fp">
  <collection name="collfp" spec="/machine/tds/data/grib/gfsfp/.*grib2$" timePartition="file"/>
  <update startup="test" />
 </featureCollection>
</dataset>

One RC dataset for each runtime, plus PofGC "collection dataset" with both runtime and time dimensions. More efficient indexing.

case 4. {collection} has multiple runtimes, time partitioned by directory: use directory partition

<dataset name="GRIB Collection Example for tutorial (directory partition)" serviceName="gribServices">
  <featureCollection name="GFS file partition" featureType="GRIB1" harvest="true" path="Grib/dp">
    <collection name="colldp" spec="/machine/tds/data/grib/rfc/**/.*grib1$" timePartition="directory"/>
    <update startup="test" />
  </featureCollection>
</dataset>

One RC dataset for each runtime, plus PofGC or PofP "collection dataset" for each directory in the tree. Scales to large collections.

case 5. {collection} has multiple runtimes, one per file, time partitioned by directory: use file partition

One RC dataset for each runtime, plus PofGC or PofP "collection dataset" for each directory in the tree. Scales to large collections. More efficient indexing.

Summary

{collection} -> {runTime} -> {Group} : a RC is a single runtime with 1 or more groups (GDS), has a single index (ncx2) file.

PofGC = {RC}

PofP = {PofGC}