Storing and updating large amount of raster data

Question

Large amount of data = ~10TB of rasters (multiple projections) with the possibility of this data growing.

The basic architecture I'm planning at the moment is having 2 clusters of GeoServer--one cluster (2 machines) as a WMS and another as a WFS. I'd like for these clusters to point to one repository for their data (raster and vector) and I am trying to figure out whether it's better to store my raster data in a file system or a database.

I've been reading "PostGIS in Action" and I'm leaning towards a DB structure for the rasters, but I'm still on the fence about it.

I have two main goals that the structure must be able to support:

1. Automate the process of building pyramids to load up onto GeoServer
2. Target specific tiles to update when new rasters come in

Right now I know I can use gdal_retile to manually build the pyramids--that coupled with GS REST API I can automate the uploading and building (although I am admittedly having trouble getting GeoServer-Manager running). But I feel a bit weary just leaving stuff in a file system given the large amount of data and the updating of files (probably going to be monthly, but could be more frequently). Because of this, the DB approach seems more appealing--but I am just a software engineer; GIS is completely new to me and I'd like some wisdom from people who do this before I start building things that can't scale properly.

Answer 1

No, do not load the imagery into PostGIS. You want to load the image catalogue into PostGIS and use the JDBC image mosaicking support in GeoServer to read the mosaics using the database catalogue. That was you can easily update which images participate in your mosaics via the catalogue in the database, while not incurring the performance penalties of storing the raw imagery itself in the database.