GanosBase uses multiple CPU cores to accelerate queries and computations on raster data through parallel operations. Two types of parallelism are supported:
Parallel SQL statements: The database distributes query execution across multiple CPU cores to scan and filter raster objects faster.
Parallel raster-level operations: GanosBase splits a single large raster object into subsets and processes them simultaneously across multiple CPU cores.
Use SQL-level parallelism to query or scan large numbers of raster objects. Use raster-level parallelism to process a single large raster object whose computation is time-consuming.
Prerequisites
Before you begin, configure the following parameters:
|
Parameter |
Description |
Default value |
Recommended value |
|
max_prepared_transactions |
Maximum number of transactions that can be in the prepared state simultaneously. |
|
Set to the value of max_connections |
|
ganos.parallel.degree |
Default degree of parallelism (DOP). A value of |
|
A positive integer based on your workload |
|
ganos.parallel.transaction |
Transactional consistency mode for parallel operations. |
|
Based on your consistency requirements |
After you modify max_prepared_transactions, you must restart your instance for the change to take effect.
max_prepared_transactions
Set max_prepared_transactions to the value of max_connections. This ensures one prepared-state transaction slot per connection. The default value is 0.
ganos.parallel.degree
The ganos.parallel.degree Grand Unified Configuration (GUC) parameter sets the default degree of parallelism (DOP). The default value is 1, which disables parallel execution.
When a function that supports parallel operations is called without an explicit DOP, or with DOP set to 0, the system uses the value of ganos.parallel.degree as the DOP.
For functions that support parallel raster-level operations, pass the DOP directly in the function call to split computation on a raster object into that number of parallel tasks:
SELECT ST_ImportFrom('chunk_table','OSS://<akxxxx>:<ak_secretxxxx>@oss-cn-beijing-internal.aliyuncs.com/mybucket/data/image.nc:hcc', '{}', '{"parallel": 4}');ganos.parallel.transaction
The ganos.parallel.transaction GUC parameter controls the transactional consistency mode for parallel operations. Valid values:
|
Value |
Description |
|
transaction_commit (default) |
Parallel transactions commit or roll back with the primary transaction. Use this mode when data consistency is required. |
|
fast_commit |
Parallel transactions cannot be rolled back. Use this mode for higher throughput when rollback support is not needed. |
Parallel SQL statements
How parallel SQL statements work
PostgreSQL generates multiple parallel query plans and distributes them across available CPU cores. Each core processes a portion of the data independently, reducing overall query execution time.
When to use parallel SQL statements
Use parallel SQL statements when scanning a large number of raster objects to identify those that meet specific spatial or attribute criteria. This reduces query execution time for read-heavy analytical workloads.
Supported scope
All read-only GanosBase functions that compute or query the attributes of raster objects support parallel queries.
Parallel raster-level operations
How parallel raster-level operations work
GanosBase divides a raster object into subsets and assigns each subset to a separate CPU core. Each subset is computed independently. After all subsets finish, the results are merged and the operation completes. This reduces the total computation time for large raster objects.
When to use raster-level parallelism
Use raster-level parallelism when processing a single large raster object whose computation would take a long time to complete serially.
Supported functions
The following functions support parallel raster-level operations.
Data import and mosaic
ST_ImportFrom — imports raster data from an external source into a chunk table
ST_MosaicFrom — merges multiple raster objects or files into a single mosaic
Pyramid building
ST_BuildPyramid — builds a multi-resolution pyramid for faster tile rendering
Geometric correction
ST_RPCRectify — applies RPC-based geometric correction to satellite imagery
Subsetting
ST_SubRaster — extracts a spatial or band-based subset from a raster object
Statistics and analysis
ST_BuildHistogram — computes the value distribution histogram for raster bands
ST_SummaryStats — returns summary statistics (min, max, mean, stddev) for raster bands
ST_BuildPercentiles — computes percentile breakpoints across raster band values
ST_ComputeStatistics — computes and stores statistical metadata for a raster object
Image processing
ST_LinearStretch — applies linear contrast stretching to raster band values
ST_InterpolateRaster — fills NoData cells using spatial interpolation
Before you begin
Create a chunk table in advance using
ST_CreateChunkTableto ensure optimal performance.If a parallel function cannot return a temporary chunk table created by an anonymous user, create the chunk table first and specify it in the chunktable parameter.