Links without nodes

[duncandewhurst]

The supply side research, surfaced several examples of networks represented as collections of line strings with either:

• no data on nodes
• with nodes represented as a separate layer and no explicit relationship between links and nodes

We'll need to consider how such networks should be treated in the standard and what guidance to provide to implementers and users.

Examples

Examples covered various network types, with some access networks containing very high numbers of links.

Access (113k links)

Middle mile & access (18k links)

interactive version (VT Data - Fiber Routes 2021 layer)

Middle mile (416 links)

interactive version

[duncandewhurst]

One option is to treat each junction as a node, with a suitable value in Node.type.

[lgs85]

One set of possible use cases we haven't yet picked up on is storing, visualising and analysing open fibre data as graph data, including with graph databases. See this set of examples from Neo4J. For this set of use cases all links would need nodes (though floating nodes with no links are ok).

I think that we have two options. One is Duncan's suggestion:

to treat each junction as a node, with a suitable value in Node.type.

This puts more work onto data processors, who would need to ender 'dummy' nodes, but requires minimal extra work on the part of data users.

The second option would be to not require links to have nodes, and require data users who want to use graph tools to create their own dummy nodes.

I don't have a particularly strong preference, but do think that if we decide to not require links to have nodes that we'd need some carefully written non-normative documentation around this.

[duncandewhurst]

I'm leaning towards requiring links to have nodes. That way, the extra work of creating 'dummy' nodes only needs to happen once, rather than each time someone wants to use the data.

That said, I think we'll need to provide non-normative documentation on creating dummy nodes in either case, whether it's for publishers or users.

We can test out approaches using the examples in the issue description.

Edit: The other consideration that is pushing to towards requiring links to have nodes is that links without nodes seems mostly to be an issue for access network data, which is not the primary focus of the standard.

[duncandewhurst]

Moving to 'Todo'. The task is to author the non-normative guidance for publishers.

[duncandewhurst]

For the Alpha, we decided not to make Link.start and Link.end required. Moving this issue to the beta milestone for further consideration.