Switching

dear community,

we're integrating a virtual chassis formed by 4 Juniper qfx, 2x5100 48 t  and 2x qfx 5100 96s.

for the 48T ones, most of the acces ports are meant to be 1G ones UTP

When configuraing the L2 somehow we found out that the only way to work out was to configure these interafces and its vlan under the "xe-" interface section  vs the "ge-" interface section. Is this right? i mean, despite the speed we need on the port, are these ones always configured under the xe-x/x/x vs the ge-?

This lead us to a problem that it seems all the servers connect just at 100Mbps half-duplex.

Somehow we could not find the way to set the xe interafce speed at 1G

Even o the show interface it says that the media is phiber vs copper

Any idea s really appreciatted.

Thanks

Gabriel.

Have you look at this link:

http://www.juniper.net/techpubs/en_US/junos14.1/topics/task/configuration/ten-gigabit-interfaces-qfx-series-cli.html#jd0e112

Are you using Juniper branded optics?  If not, you need to get at least one and test with that.  Is auto-neg enabled?  You can not set the speed (see the link), but duplex on EX4600 should be Full always.

Maybe output of show chassis hardware,show interface terse and show chassis interface extensive would help.  For the last two you could do this for one or two of the interafces.

Hi

yes, we did check that info; in fact i believe only the link-mode is allowed to be changed for the 1Gb option, but not the speed. these ports are meant to be 3-speed 100/1G/10G

we're not using any optics on the access ports as these ones are the built-in copper ones, that's one of the things that's really conffussing on the cli outputs.

I'll try to provide the cli outputs

Thanks for your reply

Cheers

I"ve not used the QFX but the ACX5k is based on this same chassis and has similar flexibility with ge/xe port options.

You do configure the port namge ge/xe to match the SFP speed inserted.  These can be configured prior to having the SFP inserted or after.

Once inserted the hardware should recognize the correct speed and configure match and allow the port to come up.  You can confirm the optics installed:

show chassis pic fpc-slot 0 pic-slot 0

Adjust the fpc and pic slot numbers as needed for your interfaces to check.  This will show the type/vendor and other details of the installed SFP.

Hi Spuluka,

that's quite confussing in these series. We read in a different thread  that in fact all the interfaces are only configured  under the xe- setion even the box creates as well the ge- section. anyhow we could not find the proper informacion on the juniper documentacion repository. L2 connectivity only started wokring properly when all the access interfaces where configured under the xe- section, even these ones were 1G end nodes ports.

There's no additional SFP added to the box. This one comes with copper bulit-in interafces 3-speed.

Cheers

Sorry, I missed that this was a T model all copper.  

Looking at the interfaces documentation for your model starting on page 41.  This confirms you understanding that all ports on this device will be configured under the xe-0/0/#  stanza.

http://www.juniper.net/techpubs/en_US/junos14.1/information-products/pathway-pages/ex4600/interfaces.pdf

The previous link a little further down discusses the changes in auto-neg for the series.

http://www.juniper.net/techpubs/en_US/junos14.1/topics/task/configuration/ten-gigabit-interfaces-qfx-series-cli.html#jd0e218

Since you are coming up half duplex, I would suggest turning off auto on this interface.

Sorry as well, as I also missed the fact that this was a 48T -;(

You should leave auto-neg enabled, and if your servers are coming up at 100M full, it would appear that is what they are negotiating to with the switch.

Documentation on what actually happens with auto-neg removed/disabled is unclear, and I am trying to find that info, but I would NOT use that setting in any production environment unless you have some external form of Server redundancy.  It is the 2nd link below, I am seeking answers for.

With 14.1X53-D30 and prior our 48T's happily negotiated 1Gb properly and successfully on xe interfaces. Upon upgrading to D35 negotiation failed completely. The solution was to ensure autonegotiate is configured.

set groups set_autoneg interfaces <xe-*> ether-options auto-negotiation
set apply-groups set_autoneg

Juniper's documentation is confusing and fails to take into account 1Gb behavior on 10Gb interfaces:

"Default
Autonegotiation is automatically enabled for Gigabit Ethernet interfaces. Autonegotation is not an option for 10-Gigabit Ethernet interfaces. No explicit action is taken after the autonegotiation is complete or if the negotiation fails."

http://www.juniper.net/documentation/en_US/junos14.1/topics/reference/configuration-statement/autonegotiation-edit-interfaces-qfx-series.html

Confusingly, Juniper also claims in the same reference: "Note: Autonegotation is not supported on QFX5100 devices."

--Paul

Did you open a ticket for this one? 

It feels like a bug that should be in the PR database.  But I don't see it there yet.  They should get that patched.

Two PRs have been filed.

I am Gabi's collegue and just made the change that smicker             suggested and now it is working fine, at 1G full duplex albeit I still see the interfaces declared like Mode-Type Fiber meanwhile I was expecting them as Mode-Type Copper .

Thansk a lot for your help!

BR

Alessandro

Looks like at some point in time auto-neg got disabled, as it should have been on by default.  Smicker's change was just to enable auto-neg.  To work as 100/1G/10G TX auto-neg must be enabled.

Awesome. That was it. Thanks a lot Paul, and you all. it seems there's no correlance between the OS and the document

BR

Gabi

FYI, I am working on updating the docs.