[e2e] 150ms - tolerable latency for quakeIII

Jim Gettys jg at pa.dec.com
Thu May 24 19:11:34 PDT 2001


The other observation is that Gamers are the ones who are first most
likely to buy broadband connections when they become available, as
they get tired of being blown away by competition with lower latencies...

But I digress: these numbers are apples and oranges.

The 30ms (sometimes 50ms) is commonly known in UI quarters...

I have first hand experience at this: early in X's history, we did
experiments on rubber banding "felt" like it was completely
attached to the mouse; that was around 30ms, and I've kept that
number in my head ever since.  At 50ms, things felt slightly rubbery.

This is *NOT* reaction time; people can't decide to do things
that fast.  It is when the time is literally imperceptible.

This also jibes with frame rates on movies required to be continuous,
which is in the 20 frame/second range.

We were trying to understand, at the time, whether we needed to put
in X server support for that sort of interaction, so were directly
concerned about exactly when things were "fast enough".

This was interesting, as on machines of the day, (Microvax), RTT 
ran around 16 ms over TCP or UNIX domain sockets.  Machines are 
quite a bit faster now. :-)

Note that this is the visual path into the brain; it may very well
be that latencies in the auditory system are higher, and so I'm not
too surprised at the idea of telephone experience being different from
UI experience.  Particularly since sound travels slowly: I betcha we're
wired to tolerate a bit of latency.
				- Jim
--
Jim Gettys
Technology and Corporate Development
Compaq Computer Corporation
jg at pa.dec.com




More information about the end2end-interest mailing list