1 Oct
2014
1 Oct
'14
9:47 a.m.
On Oct 1, 2014, at 11:44 AM, Brian L. Stuart <blstuart at bellsouth.net> wrote:
That was certainly the intent, and I've certainly read worse standards
than USB, but IMHO, they failed to achieve what they intended. USB
ended up with a protocol that is much more complex and ugly than
Ethernet and required more complex silicon.
There certainly are few protocols as simple as Ethernet.
The essence of Ethernet is the frame formats. WiFi carries those over a completely
different physical layer, but since the same packets apply, 802.11 succeeded because of
Ethernet. Conversely, 802.4 and 802.5 failed in part by being different than Ethernet.
(Excessive and useless complexity is another component, of course.)
The various flavors of Ethernet are not all that similar. For each speed grade, the
physical layer and bit level encoding is completely different. Gigabit does away with
half duplex (for practical purposes) and 10G kills it completely ? but realistically, half
duplex has been obsolete for many years now.
A full duplex Ethernet MAC is a very simple beast. A matching PHY for modest speeds (100
Mb/s or so), a few meters, and common ground, is also very simple indeed. For example,
the magnetics come from the goal of supporting isolated grounds; if you don?t need that,
which a USB-like application doesn?t, then you can do without them just as USB does.
This would make a good EE project, come to think of it. Build an FPGA that uses the
Ethernet framing but with a PHY optimized for USB-like settings, and discuss what it would
cost in high volumes. Any EE 201 profs around here who would like to try that?
paul
Both of those would be different now if Ethernet had
won over USB. If
Ethernet were used for the things USB is now used for, PoE would be
ubiquitous instead of niche, Ethernet would be dirt cheap because it's made
in trillions of units, and USB, if it still existed at all, would be
expensive because it would be a low-volume niche\ thing.
USB was *designed* to be an extremely low-cost interface exactly so it could be
high-volume. For Ethernet to have "won over" USB, it
would have had to be
redesigned so that it was as naff as USB, at which point
everybody loses.
I'm not so sure about that. It appears to me that one could design
a lower cost physical interface for Ethernet that didn't require all
the magnetics by requiring that anything more than one-to-one had
to go through a switch instead of a hub or a bus. That would also
appear to simplify the MAC silicon. Now one could argue whether
that's really still Ethernet without CSMA/CD, and quite frankly, I
don't know that I care. What would make me happy about that is
that as far as controller and driver design go, the salient feature
is that we still have 802.3 frames on the wire. Then I can do
whatever I want with those frames, just like I can with Ethernet.
I'm not forced into doing something horribly complex to accomplish
something simple.
It's not the physical layer of USB that I object to (except of course
that the name doesn't match the physical characteristics). It's
everything in all the layers above the physical that makes me want
to wretch every time I find myself working with it.
BLS