Large relations slow to show - caching a solution?

The preview maps use the editing API to download all objects referenced by the relation. There are several problems with that:

• The API is meant for editing not browsing; the preview maps were a nice feature when they were introduced but being ever more heavily used they start to put undue strain on the servers. The relation/full call required will often time out for complex relations.
• Because the API is for editing, it returns much more information than strictly needed for drawing a preview map (for example, all tags and metadata like last edit date for all nodes and ways that form the relation). All this information has to be fetched from the database, put into XML form, sent over the wire, decoded by the browser, and prepared for display in OpenLayers.
• Even the largest relations are transmitted in full resolution, which may mean tens of thousands of coordinate pairs. This is of course overkill for a small preview map.

There are many ways how this could be fixed. A simple caching of the XML response would ease the strain on the database but not solve the issue of a ton of unnecessary data being transmitted to the browser. An ideal solution would probably store the relation as a pure string of coordinates, with simplified geometry, and hand that out to the browser in KML form or so. The challenge for any caching solution would be to find out when to invalidate the cache.

All this could easily be implemented by a third party relying on diff updates loaded from the OSM server - a kind of "fast relation viewer" service.

I'm sure people would appreciate if someone built such a system they could then use. As is often the case in OSM, there's no lack of good ideas, just a lack of good people to put them into practice.