Measured throughput clarification

[robwalch]

The definition for Measured throughput (mtp) states that "it SHOULD be the value used to make switching decisions". This can be at odds with "If the client is connected to multiple servers concurrently, it must take care to report only the throughput measured against the receiving server.".

Do we want to report on the value clients use to make decisions or do we need clients to provide data that can be used to compare to a host CDNs throughput measurements?

Currently, hls.js, shaka-player, and perhaps other clients report mtp based on their bandwidth estimate which takes no such care and includes application settings in its aggregate. Is this data useful or misleading?

[acbegen]

Either info is useful in its own way, but when reporting mtp, the value against that "receiving" server should be reported - not the aggregate.

[robwalch]

A "SHOULD" or "MUST", rather than "must take care to", would help clarify whether clients like hls.js and shaka-player go on sending an aggregate estimate for mtu. They could stop sending mtu until something is implemented to meet the spec requirements (like partitioning estimates by host and using a start value based on answers to #100). Or they could continue, as in most cases, the aggregate estimate will be identical or close enough.