Subnet

The subnet is likely the single most important decision that determines how an engine actually works. It’s unlikely that the entire network of all items, all sites, all enterprises, for all time can fit into the memory of one process. The problem must be broken down. Those pieces are called subnets (a subset of the network).

The key feature of a subnet is that it must be distinct from other subnets. Subnets cannot overlap. Overlapping subnets will cause a collision if they are processed out of order. They must fully cover a network. If the subnets leave out part of the network, then that part is not processed.

Subnets must be large enough to provide all the information an engine needs to be able to make proper decisions. If a subnet is too small, then it will be missing information vital to the decision leading to sub-optimal results.

For example, an engine is detecting a trend in actuals relative to forecast to determine if a forecast needs to be adjusted to compensate. A forecast record's natural key might be item, site, and date. A subnet of site and item would mean the engine could not detect a trend over a region (many sites). A subnet of all items would be too broad if there are 1,000,000 items.