Nexus 1000v strange interface counters

[PetrHavlicek]

Hello,

We noticed really strange behavior of interface counters.

On Port-Channel where are physical interfaces of ESXi host we can see this:

  300 seconds input rate 52649472 bits/sec, 27102 packets/sec
  300 seconds output rate 240516680 bits/sec, 6624 packets/sec

Members ports showing this:

  5 minute input rate 41234288 bits/second, 20364 packets/second
  5 minute output rate 221672688 bits/second, 5962 packets/second

  5 minute input rate 6786880 bits/second, 5269 packets/second
  5 minute output rate 550336 bits/second, 164 packets/second

This is same for all port-channels for ESXi hosts. On upstream switch we can see like few(less than 10) Mbps of traffic.

Environment:

vSphere 5.5U2 with all updates from VUM

Nexus 1000V in HA version 4.2(1)SV2(2.3) and version 5.2(1)SV3(1.2)

Always VEM and VSM in same version.

Have anyone same measuring? Its problem somewhere or just missimplemented counters?

Best regards,

PetrH

[Marko Pribanic]

Hi,

I have noticed the same problem in my setup, both on physical & vethernet ports:

5 minute input rate 276446070216 bits/second, 695194 packets/second
5 minute output rate 266496680 bits/second, 504669 packets/second

NX-OS version 5.2(1)SV3(1.2).

​I have tried clearing counters, but it looks like this has no effect on input/output counters, because the same numbers persist after applying the command.  

This looks like a bug, can anybody conform this?

[PetrHavlicek]

Hello Marko,

are you using this version of N1K for long period or you just testing it? Have you seen any impact on enviroment?

BTW I try move configuration to 4.2(1)SV2(2.2) and there is counter right.

  300 seconds input rate 101472 bits/sec, 93 packets/sec
  300 seconds output rate 17688 bits/sec, 5 packets/sec

Best Regards,

PetrH

[Marko Pribanic]

Hi,

I am just testing.

In my test environment there is no performance impact with 5.2(1)SV3(1.2) version.

My NIC card in the server is Intel I350, but that should not matter because the counter problem is also observed on Vethernet ports.

I noticed if I leave all of the VM's idle the counters would eventually clear out

[heiki saaver]

Same problem here with N1kV version 5.2(1)SV3(1.2).

5 minute input rate 1951392 bits/second, 3233 packets/second

5 minute output rate 61691180064 bits/second, 5103193 packets/second

Actual traffic is 500-600Mbit/s, but the output shows 60Gbit/s :D

[PetrHavlicek]

Hello heiki,

There is more funny stuff to happened.

Example:

Ethernet3/3 is up
  Port-Profile is uplink
  MTU 1500 bytes
  Encapsulation ARPA
  Port mode is trunk
  full-duplex, 20 Gb/s
  5 minute input rate 42428468464 bits/second, 3738607 packets/second
  5 minute output rate 3772211040 bits/second, 1381985 packets/second

At 20 Gb/s full-duplex interface it measured 5 average traffic 42/3 Gbps. I've seen almost 100 Gbps uniderectional traffic at one 20 Gb/s interface.

This is really problem only for 5 min counters. Others counter increment more sense-fully.

So make your own opinion. We decide ignore 5 min counters values.

Best Regards,

Petr Havlíček

[ntst]

All, was having issues with the dropped packets counters incrementing for no reason.  TAC claims this is the same bug as you were seeing.  I was instructed to try the latest 1000V release and it has resolved the dropped packet counter issue. Can you all test this version to see if your specific issues are resolved and reply back?

version 5.2(1)SV3(1.15) released near Feb 22 2016

Regards,

Carlo