[c-nsp] autonegotiating hub was: Deferred packets on 2950 10/full interface

[Michael K. Smith]

On 1/15/05 5:14 AM, "Mikael Abrahamsson" <swmike at swm.pp.se> wrote:

> On Sat, 15 Jan 2005 lee.e.rian at census.gov wrote:
> 
>> Cheap combination of both - they act like there's an internal 2 port
>> switch for switching traffic between the 10Mb hub ports and the 100Mb
>> hub ports.
> 
> But the author referred to them as being "full duplex":
> 
>>> autonegotiating or not and often assumed autonegotiating hubs
>>> really wern't autonegotiating and would go to half duplex.  Meanwhile
>>> the hub would assume the Tlan card was full duplex and setup for it.
> 
> For me a hub cannot be full duplex, whatever speed. A hub is by
> definition a half duplex device, even if it's dual speed.

Here is a pretty good resource for Clause 28 of Draft 4 of the IEEE 802.3u
working group.  http://www.scyld.com/NWay.html  This covers 10/100 ethernet;
1000 Base is a separate group (802.3z?)

There is a Technology Ability Field bit that allows you to tell the other
side what you support vis-à-vis speed and duplex.  So, it is possible to
have a hub that can have the appropriate bit set to tell the other side that
it supports 100TX, but not have a bit set for supporting 100-Full Duplex.

I think the Auto-to-Hard Set showing up as half duplex is a result of the
Parallel Detection part of the spec.  Basically, it allows for
pre-autonegotation devices to connect to a autonegotiation device, using a
Link Monitor function to determine whether or not the devices can speak to
one another.  Bottom line, it would see 100 Meg capability, but no
autonegotiation capabilities so it would have to assume Half Duplex.

Mike