Take the 2-minute tour ×
Server Fault is a question and answer site for professional system and network administrators. It's 100% free, no registration required.

I have a weird thing happening on our network at my office which I can't quite get my head around. In particular I can't tell if it's a problem with a switch, or a problem with configuration.

We have a Cisco SG300-52 switch (sw01) in the top of a rack in our server room, connected to another SG300-28 that acts as our core switch (core01). Both run layer 2 only, our firewalls do routing between VLANs. They have a dozen or so VLANs between them. Gi1 on sw01 is a trunk port connected to gi1 on core01.

(Disclosure: There are other switches in our environment but I'm pretty sure I've isolated the problem down to these two. Happy to provide more info if necessary.)

The behaviour I'm seeing is limited to one VLAN, vlan 12 -- or, at least, it's not happening on the other ones I checked (It's hard to guarantee the absence of packets), and it is: sw01 is forwarding, to core01, traffic which is between two hosts which are both plugged into sw01. (I noticed this because the IDS in our firewall gave a false positive on traffic which should not reach the firewall.)

We noticed this mostly between our two dhcp/dns servers, net01 (10.12.0.10) and net02 (10.12.0.11). net01 is physical hardware and net02 is on a VMware ESX server. net01 is connected to gi44 on sw01 and net02's ESX server to gi11.

[net01]----gi44-[sw01]-gi1----gi1-[core01]
[net02]----gi11/

Let's see some interfaces! Remember, vlan 12 is the problem vlan. Of the others I explicitly verified that vlan 27 was not affected.

Here's the two hosts' ports: esx01 contains net02.

sw01#sh run int gi11
interface gigabitethernet11
 description esx01
 lldp med disable
 switchport trunk allowed vlan add 5-7,11-13,100
 switchport trunk native vlan 27
!
sw01#sh run int gi44
interface gigabitethernet44
 description net01-1
 lldp med disable
 switchport mode access
 switchport access vlan 12
!

Here's the trunk on sw01.

sw01#sh run int gi1 
interface gigabitethernet1
 description "trunk to core01"
 lldp med disable
 switchport trunk allowed vlan add 4-7,11-13,27,100
!

And the other end of the trunk on core01.

interface gigabitethernet1
 description sw01
 macro description switch
 switchport trunk allowed vlan add 2-7,11-16,27,100
!

I have a monitor port on core01, thus:

core01#sh run int gi12
 interface gigabitethernet12
 description "monitor port"
 port monitor GigabitEthernet 1
!

And the monitor port on core01 sees unicast traffic going between net01 and net02, both of which are on sw01! I've verified this with a monitor port on sw01 that sees the net01-net02 unicast traffic leaving via gi1 too.

sw01 knows that both of those hosts are on ports that are not its trunk port:

:) ratchet$ arp -a | grep net
net02.2ndsiteinc.com (10.12.0.11) at 00:0C:29:1A:66:15 [ether] on eth0
net01.2ndsiteinc.com (10.12.0.10) at 00:11:43:D8:9F:94 [ether] on eth0

sw01#sh mac addr addr 00:0C:29:1A:66:15
Aging time is 300 sec

  Vlan        Mac Address         Port       Type    
-------- --------------------- ---------- ---------- 
   12      00:0c:29:1a:66:15      gi11     dynamic   

sw01#sh mac addr addr 00:11:43:D8:9F:94
Aging time is 300 sec

  Vlan        Mac Address         Port       Type    
-------- --------------------- ---------- ---------- 
   12      00:11:43:d8:9f:94      gi44     dynamic   

I also brought up an unused port on sw01 on vlan 12, but the unicast traffic was (as best as I could tell) not coming out that port. So it doesn't look like sw01 is pushing it out all its ports, just the right ports and also gi1!

I've verified that sw01 is not filling up its address-table:

sw01#sh mac addr count
This may take some time.
Capacity : 8192 
Free     : 7983 
Used     : 208 

The full configs for both core01 and sw01 are available: core01, sw01.

Finally, versions:

sw01#sh ver
SW version    1.1.2.0 ( date  12-Nov-2011 time  23:34:26 )
Boot version    1.0.0.4 ( date  08-Apr-2010 time  16:37:57 )
HW version    V01

core01#sh ver
SW version    1.1.2.0 ( date  12-Nov-2011 time  23:34:26 )
Boot version    1.1.0.6 ( date  11-May-2011 time  18:31:00 )
HW version    V01

So my understanding is this: sw01 should take unicast traffic for net01 and send it only out net02's port, and vice versa; none of it should go out sw01's uplink. But core01, receiving traffic on gi1 for a host it knows is on gi1, is right in sending it out all of its ports. (That is: sw01 is misbehaving, but core01 is doing what it should given the circumstances.)

My question is: Why is sw01 sending that unicast traffic out its uplink, gi1?

(And pre-emptively: yes, I know SG300s leave much to be desired, and yes, we should have spanning-tree enabled, but that's where I'm at right now.)

share|improve this question
    
Have you taken packet captures on the monitor port? Or on the servers? Its a strange issue and the only thing that I can think of to explain it the flooding mode switches enter when the mac tables fills up but as you mentioned its far from full. Whats the processor usage like? Can you post a .pcap file from the monitor port of the traffic in question? is it only DNS and DHCP traffic that is seen on the uplink? –  GerryEgan Jun 26 '13 at 18:59
    
Do you use NLB or failover clustering ? –  Mathieu Chateau Jul 31 '13 at 15:47
add comment

1 Answer

I would take a look at the mac addresses in the captured packets. If the mask, arp table, or gateway on one of the hosts is wrong it could be setting the wrong destination mac address.

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.