EtherChannel Trunk problem

nawas · ‎09-27-2007

I have several EtherChannel trunk bundled with two fiber in our 6509s. I have always noticed that if one link goes there is always a hiccup in this netwrok, for example we have two building connected with Etherchannel trunk and this moring one trunk link went down, I saw a trunk up and down syslog message and then the whole building was offline for almost a min.

Is there anything can be done to avoid this?

Kevin Dorrell · ‎09-27-2007

Firstly, are you sure they are bundled OK? Can you post a show etherchannel from when it is good?

Secondly, is this a direct connection, or does it go through other equipment? That is, when one end goes down, does the other end get to know about it right away.

Third, if you are getting this sort of problem, you might consider forcing trunk and etherchannel status rather than negotiating them. But we will see that from the show etherchannel and show trunk.

Do you have rapid-pvst?

I suspect what is happening is that the channel is unbundling for some reason, but the links are establishing seperately, forcing a spanning-tree reconvergence, which takes 50 seconds if you are using the old Spanning Tree 802.1D.

Or perhaps, when one link of the bundle goes down, the Spanning-Tree port cost changes forcing a re-convergence. You can fix that by configuring a fixed cost on the link.

Or even due to an apparent loss of the root bridge.

There are some ideas. Over to you ...

Kevin Dorrell

Luxembourg

nawas · ‎09-27-2007

yes they are bundled and using PAgp. It is a direct connection between two 6500 and nothing in between. show output is attached and thank you for help. By the way we are running CatOs, 8.5.2 and 10.2(18)SXF4 on the MSFC.

Switch2 (enable) sh spantree summary

Spanning tree mode: RAPID-PVST+

Switch2> (enable) sh port channel

Port Status Channel Admin Ch

Mode Group Id

----- ---------- -------------------- ----- -----

1/2 connected desirable silent 32 1670

2/15 connected desirable silent 32 1670

---- ------------------------------- -------------------------

1/2 serial#(switch1) 2/5 WS-C6509

2/15 serial#(switch1) 1/15 WS-C6509

Switch2> (enable) sh trunk

* - indicates vtp domain mismatch

# - indicates dot1q-all-tagged enabled on the port

$ - indicates non-default dot1q-ethertype value

Port Mode Encapsulation Status Native vlan

-------- ----------- ------------- ------------ -----------

1/2 desirable dot1q trunking 1

2/15 desirable dot1q trunking 1

glen.grant · ‎09-27-2007

I think generally between cisco devices it is recommended to use the "non-silent" parameter as opposed to the silent , whether or not this would help I don't know . It should not pull down the trunk or etherchannel it should just fALL over to the remaining links in the channel group.

Kevin Dorrell · ‎09-27-2007

When a link leaves a bundle, does the STP cost of the bundle change?

KJD