ping from nexus 5548 to catalyst 3750x with packet loss

[jackson.ku]

Hi,

The following is our network diagram :

L2 switch A - Nexus 5548 ( IP routing ) --- 100Mbps FTTB --- Catalyst 3750X ( IP routing ) - Catalyst 6509 ( IP routing ) - L2 switch B

We used ping to verify the network connectivity. The ping paramater : repeat 1000 times, datagram size 1500. The ping  test result :

L2 switch A ping to L2 switch B : no packet loss

L2 switch B ping to L2 switch A : no packet loss

Catalyst 3750X ping to Nexus 5548 : no packet loss

Nexus 5548 ping to Catalyat 3750 : always some packet loss !!

We checked all interconnect interface of these switchs and no interface error.

We know the Nexus 5548 have CoPP, and it have default CoPP parameter. The CoPP parameter about ICMP :

class copp-system-class-icmp-echo
      police cir 64 kbps bc 3600000 bytes 

And I execute "show policy-map interface control-plane" to check the status for class-map copp-system-class-icmp-echo, the transmit violated is 0 bytes...

Please help to give me some advise why ping from Nexus 5548 to Catalyat 3750 have packet loss? 

Best Regards,

[Steve Fuller]

Hi,

With Cisco NX-OS the value you specify with the packet-size option is the payload size. This is different to IOS where the value is the packet size.

When you send the pings from the Nexus, you're specifying 1500-bytes which with the ICMP and IP headers, results in a packet size of 1528-bytes. Unless you have jumbo frames enabled these packets are then being fragmented before being sent.

Can you try ping <3750-ip-address> count 1000 packet-size 1472 df-bit and see if you have the same issue?

Regards

[jackson.ku]

HI,

You are correct. The ping was ok If we ping with size 1472.

But the ping packet not always lost if datagram size is 1500 ( ping 1000 times, loss about 10 times )?? 

Best Regards,

[Steve Fuller]

Hi,

I don't know any commands that can be used to positively confirm this, but I suspect the small loss you're now seeing is due to the additional effort for fragmentation and re-assembly of the large packets.

You could check the output of show process cpu | ex 0.0 to check the CPU utilisation of the Catalyst switch when the Nexus is pinging it, and maybe the show ip traffic to see if any of those counters increase when there's a ping drop, but I don't think either command will give you any definite results.

Given the hosts attached to the downstream Layer-2 switches should not send traffic that will require fragmentation, I think it's probably safe to consider the link and network connectivity between them as "clean".

Regards