`removeLatencyFromFirstPacketTime: true` causes time to first byte to be
excluded from throughput measurements, greatly inflating bandwidth
estimates by as much as 2x. This causes elevated rebuffering rates.
Measurements were taken on various Cast models, as well as Chrome on
Linux. In all cases, bandwidth estimates after 60 seconds of 4K Sintel
with the flag set to `true` were about 2x what they would be with the
flag set to `false`.
It appears that time to first byte is excluded from all requests, rather
than just the first segment as is implied by the config docs. It's also
unclear to me, at least, what this was supposed to accomplish, even if
only applied to one segment. For now I'm neither removing nor attempting
to fix the feature, until we can discuss the original thinking behind
it.
As an emergency fix, because the default behavior was so problematic,
this changes the default to `false`. Applications that know better than
me still have the option to set it explicitly to `true` for now.
See issue #9475, which will stay open until we decide what to do with
this flag and its implementation.
Joey Parrish
a84c6b53c3