From: Rob (rob@warnerbeach.com)
Date: Mon Feb 27 2006 - 15:17:41 GMT-3
Hi All,
Interesting problem Id like to share and if anyone's ever seen anything like
it would appreciate feedback.
I know its not necessarily CCIE topic material but it may help in some small
way
I have the following scenario
Pc---10.1.1.10/24-----29050switch-defaultvlan
-ip10.1.1.7/24---d/c-gig-3550-vlan14-ip 10.1.1.1/24-6directlyconnectednets+
L3gigPort-10.20.1.1/30----------gig---6500-----10.20.1.2/30
The 3750 has an eigrp relationship with the 6500 and has a default gateway
out to 10.1.1.5/24. The 2950 has a default gateway of 10.1.1.7 and the 6500
knows all of the routes via eigrp. There are other devices but for this
scenario they don't matter.
I can ping out from the pc to any net directly connected to the 3550 but if
I go to any network beyond it fails(gateways of last resort are set)
I have 20 devices in the same vlan as the pc and it only ever fails on one
ip address outside of the vlan. Inside the vlan its always visible.
The problem moves around in the same vlan from device to device but never
more than one ip address at a time.
The gateways are all set correctly and Ive disabled icmp redirects on all
interfaces.
I put acls on to try and find out where the packet was failing
From the 6500 I generated icmp packets and saw the request hit the 3550 ,
route through the gig uplink , over to the SVI and then out over the
physical link towards the 2950..
On the 2950 and sniffing on the PC I never saw the request. Likewise
generating traffic from the PC out I never saw it get to the 3550
Now if I give the pc a different address on the same subnet it suddenly
starts working. Likewise if I clear the arp-cache or the routing table on
the 3550 the problem goes away for a day or two then re-surfaces.
During my tests Ive seen all arp entries are good , there are no mac-port
failures, the routing tables are stable , the mac tables accurate, default
geways are set , link status is up/up. I realise the vlan1 configuration is
not the optimum but this isn't my network. I have recommended the native
vlan be changed at the access-layer but this isn't the cause of the problem.
All of my tests point towards the 2950 as there are another 3 2950s off the
same 3550 and if I move the device that fails into one of those switchs the
problem goes away.If I put it back the problem re-appears until the
arp-cache or routing table on 3550 gets flushed! This is what is stumping me
at the moment
Im thinking IOS bug but have never seen anything quite like it. Could I be
missing something obvious?
During the sniff I did find the 3550 sending icmp redirects for the icmp
request to go via 10.1.1.5 so disabled them although the problem still
currently exists.
TIA
This archive was generated by hypermail 2.1.4 : Wed Mar 01 2006 - 11:28:18 GMT-3