[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Ecrit] LoST civic caching issue



Hi Henning,

thanks for raising this issue and it is in fact a difficult problem.

I see two aspects with the "the donut problem", i.e., there is a larger service boundary with another service area within (the hole in the donut). There are two cases:

* The entity responsible for the larger service area knows that there is a donut problem. In this case it would be reasonable to assume that this entity could mark the response as "non cachable". For geo one would want to express the service area as a donut -- currently we don't have a description how to describe such a shape. I now believe that we should add something to the PIDF-LO profile draft.

* The entity responsible for the larger service area does NOT know about a potential donut problem. This case is actually not related to civic since it can appear in geo as well. I don't have an easy and nice answer for this case.

Ciao
Hannes

Henning Schulzrinne wrote:
During our implementation of caching in LoST, one of my students came up with the following interesting issue:

Suppose that a system has cached a mapping for city=New York, state=NY and somebody moves the client to a place within that region that has its own mapping, e.g., a college campus. (Similar issues arise for places where a county has one mapping, with some part, e.g., the incorporated area, having a more specific mapping.) In those cases, a client would notice that the address has changed, but it still matches the old mapping, so it will not re-query.

A related problem occurs if, during a failure, a default response has been returned.

For geo coordinates, 'donuts' and other ways to express 'all but this area' conditions, which would trigger re-querying. This doesn't work for civic coordinates, obviously.

There seem to be several possibilities:

(1) Don't do that. In other words, if an area knows that there are more specific mappings, it has to be split up in some way, e.g., by enumerating all such areas explicitly. For example, for the unincorporated areas, the community name is part of the mapping, even though they all share the same county-wide mapping.

This probably works for most cases, but fails for the campus example, since the surrounding city doesn't really know about these.

(2) Mark such cases as non-cacheable. Since civic coordinates are only used for stationary and nomadic devices, the additional query load is rather modest. This is the safe approach.

(3) As a compromise between (1) and (2), mark mappings that are known to be leaf nodes. For example, a campus could mark its mapping as such since it can be pretty sure that there's nothing below that level. That way, laptops moving within a building don't have to requery. These could simply be marked as cacheable, so that no additional XML tags are needed.

Henning



_______________________________________________
Ecrit mailing list
Ecrit at ietf.org
https://www1.ietf.org/mailman/listinfo/ecrit


_______________________________________________
Ecrit mailing list
Ecrit at ietf.org
https://www1.ietf.org/mailman/listinfo/ecrit