Peter upgraded the stream operator for spatial interpolation. The new version of this operator uses in-operator memory more efficiently, which allows multiple operators to run in parallel. Peter also added an option to the operator which controls how the grid positions are loaded and saved. If the number of the grid positions is not too large, they can be loaded and saved in the in-operator memory before the scattered data are received. Once all the scattered data are received (a “flush” event of a stream’s tumbling window), the operator can begin interpolating the surface immediately. Otherwise, data can be saved in persistent memory (a file or a database) and be queried after the scattered data are all received. This improved operator uses less memory, at the cost of increased latency, spending more time waiting for grid positions.