On 2/11/2025 11:39 PM, to...@strayalpha.com wrote:
In a later post you assess that 13ms reordering window acceptable and in
accordance with the above recommendations, so let me ask, what is the threshold
for acceptable and non-acceptable (max) reordering-induced delays?
My metric is that delay increase matter only when they are a significant
fraction of the current message latency. 13ms probably under 25% of most
Internet delays, if not lower. It’s also a small fraction (about 1/6) of the
delay a human would notice (100ms). So its impact is small both relatively
(25%) and as an absolute (1/6 of notiicible).
I think the modern latency numbers are much smaller. RTT of connections
going though a CDN is typically lower than 35ms, and in many cases lower
than 15ms. Adding 13ms to that is quite a big penalty.
The delay that humans can notice is largely depending of task and
context. The often cited figure of 150ms relates to turn taking in
natural audio conversation, when silence after saying a sentence is the
main clue that the current speaker has stopped speaking. If the latency
is greater than 150ms, most people notice it and the communication
becomes awkward. There are other important figures, like a latency of
10ms for hand-eye coordination, which is very relevant in video-games .
There are even stronger requirements for virtual reality, when even a
few milliseconds between detecting an eye movement and "repainting" the
screen . Adding 13 ms would be very noticeable in video games, and would
be a complete non starter for remote virtual reality displays.
Then there is machine communication. Displaying a web page, for example,
may well require series of transactions to fetch data from separate
databases. compose images, etc. Adding 13ms to several of these
transactions will have a cumulative effect, very noticeable by humans.
Modern Internet application do require low delays. Gratuitously adding
delays "because we always did delay correction at L2" is a very bad idea.
-- Christian Huitema
