hard-coding full-duplex

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

I believe the spec says if one port is hard-coded full-duplex
and the other isn't, then that port negotiates half-duplex.
Is this correct ? Pointers to any RFC's or proof appreciated.

The question is: what happens on a VLAN on a switch with four
ports but only one is hard-coded full duplex? Would the other
3 ports be forced to half?
Would any other ports on the switch on different VLANs be affected ?

alan

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <cbrtqo$rb4$1@internal.wj.com>,
alan@internal.wj.com (Alan Strassberg) wrote:

> I believe the spec says if one port is hard-coded full-duplex
> and the other isn't, then that port negotiates half-duplex.
> Is this correct ? Pointers to any RFC's or proof appreciated.
>

If one side of a 10/100BASE-T link is manually configured for
full-duplex operation (by disabling Auto-Negotiation), and the other
side of the link is allowed to Auto-Negotiate, the negotiating side will
indeed resolve to half-duplex operation. This happens because the
negotiating side will "parallel detect" its link partner as operating in
either 10 or 100BASE-T. However, there is no way to determine that a
device is configured for full duplex absent some advertisement
(typically via Auto-Negotiation). Lacking any knowledge of the
duplexity, the "safest" route is to assume half-duplex operation.

This algorithm is defined in IEEE 802.3 (Clause 28).

"Hard coding" for full-duplex can also be achieved by forcing
Auto-Negotiation to advertise that the device can only (or is only
willing to) operate in full-duplex mode (as opposed to disabling A-N
altogether). This is a preferable method of forcing full-duplex
operation, since a negotiating link partner will resolve to full-duplex
operation (assuming it is capable of doing so), and thus avoid the
duplex mismatch.

In Gigabit Ethernet (1000BASE-T) Auto-Negotiation is not optional; i.e.,
it cannot be disabled as in 10/100BASE-T. One side of a Gigabit Ethernet
link can be forced to full-duplex operation only by forcing A-N to
advertise that it is only willing (or capable) of operating in
full-duplex mode; indeed, this is commonly done. A negotiating link
partner will see this advertisement and configure itself for full duplex
mode in response, resulting in a properly operating full-duplex link.


> The question is: what happens on a VLAN on a switch with four
> ports but only one is hard-coded full duplex? Would the other
> 3 ports be forced to half?
> Would any other ports on the switch on different VLANs be affected ?
>

Auto-Negotiation is performed on a link-by-link basis; the link
configuration on one port of a switch has no effect on the configuration
of other ports on that same switch. Note that this has nothing to do
with VLANs at all; the logical memberships of the ports on a switch with
respect to Virtual LANs is completely unrelated to the duplex
configuration of the ports.


--
Rich Seifert Networks and Communications Consulting
21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95033
(408) 395-1966 FAX

Send replies to: usenet at richseifert dot com

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

"Rich Seifert" <usenet-@-richseifert-dot-com.invalid> wrote in message
news:usenet--1C18C9.07594429062004@news-central.dca.giganews.com...
> In article <cbrtqo$rb4$1@internal.wj.com>,
> alan@internal.wj.com (Alan Strassberg) wrote:
>
> > I believe the spec says if one port is hard-coded full-duplex
> > and the other isn't, then that port negotiates half-duplex.
> > Is this correct ? Pointers to any RFC's or proof appreciated.
> >
>
> If one side of a 10/100BASE-T link is manually configured for
> full-duplex operation (by disabling Auto-Negotiation), and the other
> side of the link is allowed to Auto-Negotiate, the negotiating side will
> indeed resolve to half-duplex operation. This happens because the
> negotiating side will "parallel detect" its link partner as operating in
> either 10 or 100BASE-T. However, there is no way to determine that a
> device is configured for full duplex absent some advertisement
> (typically via Auto-Negotiation). Lacking any knowledge of the
> duplexity, the "safest" route is to assume half-duplex operation.
>
> This algorithm is defined in IEEE 802.3 (Clause 28).
>
> "Hard coding" for full-duplex can also be achieved by forcing
> Auto-Negotiation to advertise that the device can only (or is only
> willing to) operate in full-duplex mode (as opposed to disabling A-N
> altogether). This is a preferable method of forcing full-duplex
> operation, since a negotiating link partner will resolve to full-duplex
> operation (assuming it is capable of doing so), and thus avoid the
> duplex mismatch.
>
> In Gigabit Ethernet (1000BASE-T) Auto-Negotiation is not optional; i.e.,
> it cannot be disabled as in 10/100BASE-T. One side of a Gigabit Ethernet
> link can be forced to full-duplex operation only by forcing A-N to
> advertise that it is only willing (or capable) of operating in
> full-duplex mode; indeed, this is commonly done. A negotiating link
> partner will see this advertisement and configure itself for full duplex
> mode in response, resulting in a properly operating full-duplex link.
>
>
> > The question is: what happens on a VLAN on a switch with four
> > ports but only one is hard-coded full duplex? Would the other
> > 3 ports be forced to half?
> > Would any other ports on the switch on different VLANs be affected ?
> >
>
> Auto-Negotiation is performed on a link-by-link basis; the link
> configuration on one port of a switch has no effect on the configuration
> of other ports on that same switch. Note that this has nothing to do
> with VLANs at all; the logical memberships of the ports on a switch with
> respect to Virtual LANs is completely unrelated to the duplex
> configuration of the ports.

Just wanted to add that ports that are configured to be in the same Fast
Etherchannel group will act as one.
If a speed or duplex change is made to one port in a 4-port Fast
Etherchannel group then all ports will convert
to the new setting. Same with a 2-port FastEtherchannel group. Also, Cisco
has a pretty good chart showing the
resulting speed/duplex of NIC cards and switch ports related to hard-coded
vs. auto-negotiate settings. Here's the
URL (see Table 1).

http://www.cisco.com/en/US/products/hw/switches/ps700/products_tech_note09186a00800a7af0.shtml

Cheers!

Cisco_Kid




>
>
> --
> Rich Seifert Networks and Communications Consulting
> 21885 Bear Creek Way
> (408) 395-5700 Los Gatos, CA 95033
> (408) 395-1966 FAX
>
> Send replies to: usenet at richseifert dot com

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

One related question from my side. Suppose i have 5 computers running at
100Mbps full duplex and then I plug in another system to the network which
is only 10 Mbps capable. would Autonegotiation be performed? if yes would
all the systems come down to 10 Mbps speed?
As I believe AN is performed between two link partners then how come all the
computers would talk to the new system at 10 Mbps


"Cisco_Kid" <cisco_kid_20000@hotmail.com> wrote in message
news:cc1f2d$e8l$1@home.itg.ti.com...
>
> "Rich Seifert" <usenet-@-richseifert-dot-com.invalid> wrote in message
> news:usenet--1C18C9.07594429062004@news-central.dca.giganews.com...
> > In article <cbrtqo$rb4$1@internal.wj.com>,
> > alan@internal.wj.com (Alan Strassberg) wrote:
> >
> > > I believe the spec says if one port is hard-coded full-duplex
> > > and the other isn't, then that port negotiates half-duplex.
> > > Is this correct ? Pointers to any RFC's or proof appreciated.
> > >
> >
> > If one side of a 10/100BASE-T link is manually configured for
> > full-duplex operation (by disabling Auto-Negotiation), and the other
> > side of the link is allowed to Auto-Negotiate, the negotiating side will
> > indeed resolve to half-duplex operation. This happens because the
> > negotiating side will "parallel detect" its link partner as operating in
> > either 10 or 100BASE-T. However, there is no way to determine that a
> > device is configured for full duplex absent some advertisement
> > (typically via Auto-Negotiation). Lacking any knowledge of the
> > duplexity, the "safest" route is to assume half-duplex operation.
> >
> > This algorithm is defined in IEEE 802.3 (Clause 28).
> >
> > "Hard coding" for full-duplex can also be achieved by forcing
> > Auto-Negotiation to advertise that the device can only (or is only
> > willing to) operate in full-duplex mode (as opposed to disabling A-N
> > altogether). This is a preferable method of forcing full-duplex
> > operation, since a negotiating link partner will resolve to full-duplex
> > operation (assuming it is capable of doing so), and thus avoid the
> > duplex mismatch.
> >
> > In Gigabit Ethernet (1000BASE-T) Auto-Negotiation is not optional; i.e.,
> > it cannot be disabled as in 10/100BASE-T. One side of a Gigabit Ethernet
> > link can be forced to full-duplex operation only by forcing A-N to
> > advertise that it is only willing (or capable) of operating in
> > full-duplex mode; indeed, this is commonly done. A negotiating link
> > partner will see this advertisement and configure itself for full duplex
> > mode in response, resulting in a properly operating full-duplex link.
> >
> >
> > > The question is: what happens on a VLAN on a switch with four
> > > ports but only one is hard-coded full duplex? Would the other
> > > 3 ports be forced to half?
> > > Would any other ports on the switch on different VLANs be affected ?
> > >
> >
> > Auto-Negotiation is performed on a link-by-link basis; the link
> > configuration on one port of a switch has no effect on the configuration
> > of other ports on that same switch. Note that this has nothing to do
> > with VLANs at all; the logical memberships of the ports on a switch with
> > respect to Virtual LANs is completely unrelated to the duplex
> > configuration of the ports.
>
> Just wanted to add that ports that are configured to be in the same Fast
> Etherchannel group will act as one.
> If a speed or duplex change is made to one port in a 4-port Fast
> Etherchannel group then all ports will convert
> to the new setting. Same with a 2-port FastEtherchannel group. Also, Cisco
> has a pretty good chart showing the
> resulting speed/duplex of NIC cards and switch ports related to hard-coded
> vs. auto-negotiate settings. Here's the
> URL (see Table 1).
>
>
http://www.cisco.com/en/US/products/hw/switches/ps700/products_tech_note0918
6a00800a7af0.shtml
>
> Cheers!
>
> Cisco_Kid
>
>
>
>
> >
> >
> > --
> > Rich Seifert Networks and Communications Consulting
> > 21885 Bear Creek Way
> > (408) 395-5700 Los Gatos, CA 95033
> > (408) 395-1966 FAX
> >
> > Send replies to: usenet at richseifert dot com
>
>

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <cc2mgc$bnp$1@newssrv.muc.infineon.com>,
zanzyinlove <zanzyinlove@zanzyland.com> wrote:
:One related question from my side. Suppose i have 5 computers running at
:100Mbps full duplex and then I plug in another system to the network which
:is only 10 Mbps capable. would Autonegotiation be performed?

Yes, unless you have disabled it. The fact that you are running at
full duplex tells us that you are using a switch rather than a hub;
when you plug in the additional device to the switch, then provided
you haven't turned off autonegotiation, it will negotiate to the
appropriate 10 Mbps setting for that one link.

:if yes would
:all the systems come down to 10 Mbps speed?

No. Switches buffer.

:As I believe AN is performed between two link partners then how come all the
:computers would talk to the new system at 10 Mbps

In each case, the link partners would be the device on one end
and the switch on the other end. The switch then buffers as necessary
so as to be able to transmit to the destination port at the appropriate
speed and duplex for that destination port. The magic is all in the switch.

Now, if you had not happpened to mention full duplex, then there would
be the possibility that you were using a repeater (hub) rather than a
switch; all ports on a repeater must operate at the same speed. As
repeaters are always half duplex, and all the ports have to go the same
speed, usually the repeater is fixed at a -particular- speed -- you
buy a 10 Mbps repeater or a 100 Mbps repeater, not a repeater that
runs at 100 if all the devices can and otherwise drops down to 10.

To be fair/accurate, some models of repeaters *do* negotiate
speed on a port-by-port basis, with anything that happens to come
in at 10 Mbps being sent only to the ports that happen to be running
at 10 Mbps, and anything that happens to come in on a 100 Mbps port
only being repeated to the other ports that happen to be at at 100 Mbps.
In the 3Com model we bought [several years ago], the "enhanced"
version also offered a bridge between the two speeds... but then
anything transitioning between the two speeds would be -bridged- instead
of being -repeated-, so really the device was operating as a switch
between the two speeds rather than as a hub..,
--
I wrote a hack in microcode,
with a goto on each line,
it runs as fast as Superman,
but not quite every time! -- Don Libes et al.

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

thanks a lot for the wonderful detailed explanation.
"Walter Roberson" <roberson@ibd.nrc-cnrc.gc.ca> wrote in message
news:cc2un4$8kj$1@canopus.cc.umanitoba.ca...
> In article <cc2mgc$bnp$1@newssrv.muc.infineon.com>,
> zanzyinlove <zanzyinlove@zanzyland.com> wrote:
> :One related question from my side. Suppose i have 5 computers running at
> :100Mbps full duplex and then I plug in another system to the network
which
> :is only 10 Mbps capable. would Autonegotiation be performed?
>
> Yes, unless you have disabled it. The fact that you are running at
> full duplex tells us that you are using a switch rather than a hub;
> when you plug in the additional device to the switch, then provided
> you haven't turned off autonegotiation, it will negotiate to the
> appropriate 10 Mbps setting for that one link.
>
> :if yes would
> :all the systems come down to 10 Mbps speed?
>
> No. Switches buffer.
>
> :As I believe AN is performed between two link partners then how come all
the
> :computers would talk to the new system at 10 Mbps
>
> In each case, the link partners would be the device on one end
> and the switch on the other end. The switch then buffers as necessary
> so as to be able to transmit to the destination port at the appropriate
> speed and duplex for that destination port. The magic is all in the
switch.
>
> Now, if you had not happpened to mention full duplex, then there would
> be the possibility that you were using a repeater (hub) rather than a
> switch; all ports on a repeater must operate at the same speed. As
> repeaters are always half duplex, and all the ports have to go the same
> speed, usually the repeater is fixed at a -particular- speed -- you
> buy a 10 Mbps repeater or a 100 Mbps repeater, not a repeater that
> runs at 100 if all the devices can and otherwise drops down to 10.
>
> To be fair/accurate, some models of repeaters *do* negotiate
> speed on a port-by-port basis, with anything that happens to come
> in at 10 Mbps being sent only to the ports that happen to be running
> at 10 Mbps, and anything that happens to come in on a 100 Mbps port
> only being repeated to the other ports that happen to be at at 100 Mbps.
> In the 3Com model we bought [several years ago], the "enhanced"
> version also offered a bridge between the two speeds... but then
> anything transitioning between the two speeds would be -bridged- instead
> of being -repeated-, so really the device was operating as a switch
> between the two speeds rather than as a hub..,
> --
> I wrote a hack in microcode,
> with a goto on each line,
> it runs as fast as Superman,
> but not quite every time! -- Don Libes et al.