Welcome to the SDP Resource Model’s documentation!
The SDP Resource Model is a Python application for simulating and visualising the data storage and compute requirements of the SDP for during batch processing, given an observation schedule.
Simulation overview
A flow chart of the simulation process is shown below (click to zoom).
Assumptions
The SDP Resource Model makes the following assumptions:
There is only one telescope available
Observations and data processing are executed in the order they are scheduled (unless the shuffle option is enabled - see here for more information)
Receive & real-time processing can be ignored
Scheduling blocks can comprise multiple scheduling block instances (see here for more information on how these are defined) - raw visibilities will accumulate in storage and batch processing will be executed after the final observation of the last instance. There is no data retention period for raw visibilities.
Data products generated by pipelines are kept in data product storage from when the pipeline finishes until a data retention period, which is specified in the scheduling block type configuration and is the same for all pipelines in the block.
Observations are delayed if there is insufficient capacity storage for raw visibilities.
Storage for pre-processed visibilities is allocated and released for each scheduling block when processing of these starts and ends.
The number of nodes allocated to an individual pipeline is fixed for the duration of the pipeline (set by its
num_nodesparameter).Pipelines with the same priority level will be executed in parallel only if resources are available.
There is no overhead in spinning up a node.
Note
The SDP Resource Model is a work in progress and is subject to change.