Re: Interface. Gauging interest.

You know every time the software group says that Ethernet isn't
deterministic, I have to make them tell me exactly what they mean.
Some think it means packet loss, some think it means you can't predict
the queue performance in switches, some think it means that the stack
performance sucks, and some people think it's because Ethernet doesn't
use synchronous clocking. Well simply put, in a closed system you can
address every issue that makes Ethernet underperform. Same goes for
USB, as you illustrated by attacking the issues as a Maker rather than
a User.

My less than clear point (I tend to make a lot of less than clear
points) is that if you use rule-of-thumb engineering, you're making a
mistake.

In this case what I mean by USB not being deterministic is that you
can't (meaning it's really hard/expensive) deliver packets without
temporal jitter. It's not designed for it. 1394, on the other hand is
designed for it.

On 5/19/06, Peter Reilley <micrio@...> wrote:
> It is wrong to say that USB is not deterministic, it is. USB frames
> operate on a strict 1 millisecond cycle. This is unlike Ethernet which
> operates on a random access mechanism.